1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * ring buffer based function tracer
4 *
5 * Copyright (C) 2007-2012 Steven Rostedt <srostedt@redhat.com>
6 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 *
8 * Originally taken from the RT patch by:
9 * Arnaldo Carvalho de Melo <acme@redhat.com>
10 *
11 * Based on code from the latency_tracer, that is:
12 * Copyright (C) 2004-2006 Ingo Molnar
13 * Copyright (C) 2004 Nadia Yvette Chambers
14 */
15 #include <linux/ring_buffer.h>
16 #include <generated/utsrelease.h>
17 #include <linux/stacktrace.h>
18 #include <linux/writeback.h>
19 #include <linux/kallsyms.h>
20 #include <linux/security.h>
21 #include <linux/seq_file.h>
22 #include <linux/irqflags.h>
23 #include <linux/debugfs.h>
24 #include <linux/tracefs.h>
25 #include <linux/pagemap.h>
26 #include <linux/hardirq.h>
27 #include <linux/linkage.h>
28 #include <linux/uaccess.h>
29 #include <linux/vmalloc.h>
30 #include <linux/ftrace.h>
31 #include <linux/module.h>
32 #include <linux/percpu.h>
33 #include <linux/splice.h>
34 #include <linux/kdebug.h>
35 #include <linux/string.h>
36 #include <linux/mount.h>
37 #include <linux/rwsem.h>
38 #include <linux/slab.h>
39 #include <linux/ctype.h>
40 #include <linux/init.h>
41 #include <linux/panic_notifier.h>
42 #include <linux/kmemleak.h>
43 #include <linux/poll.h>
44 #include <linux/nmi.h>
45 #include <linux/fs.h>
46 #include <linux/trace.h>
47 #include <linux/sched/clock.h>
48 #include <linux/sched/rt.h>
49 #include <linux/fsnotify.h>
50 #include <linux/irq_work.h>
51 #include <linux/workqueue.h>
52
53 #include <asm/setup.h> /* COMMAND_LINE_SIZE */
54
55 #include "trace.h"
56 #include "trace_output.h"
57
58 #ifdef CONFIG_FTRACE_STARTUP_TEST
59 /*
60 * We need to change this state when a selftest is running.
61 * A selftest will lurk into the ring-buffer to count the
62 * entries inserted during the selftest although some concurrent
63 * insertions into the ring-buffer such as trace_printk could occurred
64 * at the same time, giving false positive or negative results.
65 */
66 static bool __read_mostly tracing_selftest_running;
67
68 /*
69 * If boot-time tracing including tracers/events via kernel cmdline
70 * is running, we do not want to run SELFTEST.
71 */
72 bool __read_mostly tracing_selftest_disabled;
73
disable_tracing_selftest(const char * reason)74 void __init disable_tracing_selftest(const char *reason)
75 {
76 if (!tracing_selftest_disabled) {
77 tracing_selftest_disabled = true;
78 pr_info("Ftrace startup test is disabled due to %s\n", reason);
79 }
80 }
81 #else
82 #define tracing_selftest_running 0
83 #define tracing_selftest_disabled 0
84 #endif
85
86 /* Pipe tracepoints to printk */
87 static struct trace_iterator *tracepoint_print_iter;
88 int tracepoint_printk;
89 static bool tracepoint_printk_stop_on_boot __initdata;
90 static DEFINE_STATIC_KEY_FALSE(tracepoint_printk_key);
91
92 /* For tracers that don't implement custom flags */
93 static struct tracer_opt dummy_tracer_opt[] = {
94 { }
95 };
96
97 static int
dummy_set_flag(struct trace_array * tr,u32 old_flags,u32 bit,int set)98 dummy_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
99 {
100 return 0;
101 }
102
103 /*
104 * To prevent the comm cache from being overwritten when no
105 * tracing is active, only save the comm when a trace event
106 * occurred.
107 */
108 static DEFINE_PER_CPU(bool, trace_taskinfo_save);
109
110 /*
111 * Kill all tracing for good (never come back).
112 * It is initialized to 1 but will turn to zero if the initialization
113 * of the tracer is successful. But that is the only place that sets
114 * this back to zero.
115 */
116 static int tracing_disabled = 1;
117
118 cpumask_var_t __read_mostly tracing_buffer_mask;
119
120 /*
121 * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
122 *
123 * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
124 * is set, then ftrace_dump is called. This will output the contents
125 * of the ftrace buffers to the console. This is very useful for
126 * capturing traces that lead to crashes and outputing it to a
127 * serial console.
128 *
129 * It is default off, but you can enable it with either specifying
130 * "ftrace_dump_on_oops" in the kernel command line, or setting
131 * /proc/sys/kernel/ftrace_dump_on_oops
132 * Set 1 if you want to dump buffers of all CPUs
133 * Set 2 if you want to dump the buffer of the CPU that triggered oops
134 */
135
136 enum ftrace_dump_mode ftrace_dump_on_oops;
137
138 /* When set, tracing will stop when a WARN*() is hit */
139 int __disable_trace_on_warning;
140
141 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
142 /* Map of enums to their values, for "eval_map" file */
143 struct trace_eval_map_head {
144 struct module *mod;
145 unsigned long length;
146 };
147
148 union trace_eval_map_item;
149
150 struct trace_eval_map_tail {
151 /*
152 * "end" is first and points to NULL as it must be different
153 * than "mod" or "eval_string"
154 */
155 union trace_eval_map_item *next;
156 const char *end; /* points to NULL */
157 };
158
159 static DEFINE_MUTEX(trace_eval_mutex);
160
161 /*
162 * The trace_eval_maps are saved in an array with two extra elements,
163 * one at the beginning, and one at the end. The beginning item contains
164 * the count of the saved maps (head.length), and the module they
165 * belong to if not built in (head.mod). The ending item contains a
166 * pointer to the next array of saved eval_map items.
167 */
168 union trace_eval_map_item {
169 struct trace_eval_map map;
170 struct trace_eval_map_head head;
171 struct trace_eval_map_tail tail;
172 };
173
174 static union trace_eval_map_item *trace_eval_maps;
175 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
176
177 int tracing_set_tracer(struct trace_array *tr, const char *buf);
178 static void ftrace_trace_userstack(struct trace_array *tr,
179 struct trace_buffer *buffer,
180 unsigned int trace_ctx);
181
182 #define MAX_TRACER_SIZE 100
183 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
184 static char *default_bootup_tracer;
185
186 static bool allocate_snapshot;
187 static bool snapshot_at_boot;
188
189 static char boot_instance_info[COMMAND_LINE_SIZE] __initdata;
190 static int boot_instance_index;
191
192 static char boot_snapshot_info[COMMAND_LINE_SIZE] __initdata;
193 static int boot_snapshot_index;
194
set_cmdline_ftrace(char * str)195 static int __init set_cmdline_ftrace(char *str)
196 {
197 strscpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
198 default_bootup_tracer = bootup_tracer_buf;
199 /* We are using ftrace early, expand it */
200 trace_set_ring_buffer_expanded(NULL);
201 return 1;
202 }
203 __setup("ftrace=", set_cmdline_ftrace);
204
set_ftrace_dump_on_oops(char * str)205 static int __init set_ftrace_dump_on_oops(char *str)
206 {
207 if (*str++ != '=' || !*str || !strcmp("1", str)) {
208 ftrace_dump_on_oops = DUMP_ALL;
209 return 1;
210 }
211
212 if (!strcmp("orig_cpu", str) || !strcmp("2", str)) {
213 ftrace_dump_on_oops = DUMP_ORIG;
214 return 1;
215 }
216
217 return 0;
218 }
219 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
220
stop_trace_on_warning(char * str)221 static int __init stop_trace_on_warning(char *str)
222 {
223 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
224 __disable_trace_on_warning = 1;
225 return 1;
226 }
227 __setup("traceoff_on_warning", stop_trace_on_warning);
228
boot_alloc_snapshot(char * str)229 static int __init boot_alloc_snapshot(char *str)
230 {
231 char *slot = boot_snapshot_info + boot_snapshot_index;
232 int left = sizeof(boot_snapshot_info) - boot_snapshot_index;
233 int ret;
234
235 if (str[0] == '=') {
236 str++;
237 if (strlen(str) >= left)
238 return -1;
239
240 ret = snprintf(slot, left, "%s\t", str);
241 boot_snapshot_index += ret;
242 } else {
243 allocate_snapshot = true;
244 /* We also need the main ring buffer expanded */
245 trace_set_ring_buffer_expanded(NULL);
246 }
247 return 1;
248 }
249 __setup("alloc_snapshot", boot_alloc_snapshot);
250
251
boot_snapshot(char * str)252 static int __init boot_snapshot(char *str)
253 {
254 snapshot_at_boot = true;
255 boot_alloc_snapshot(str);
256 return 1;
257 }
258 __setup("ftrace_boot_snapshot", boot_snapshot);
259
260
boot_instance(char * str)261 static int __init boot_instance(char *str)
262 {
263 char *slot = boot_instance_info + boot_instance_index;
264 int left = sizeof(boot_instance_info) - boot_instance_index;
265 int ret;
266
267 if (strlen(str) >= left)
268 return -1;
269
270 ret = snprintf(slot, left, "%s\t", str);
271 boot_instance_index += ret;
272
273 return 1;
274 }
275 __setup("trace_instance=", boot_instance);
276
277
278 static char trace_boot_options_buf[MAX_TRACER_SIZE] __initdata;
279
set_trace_boot_options(char * str)280 static int __init set_trace_boot_options(char *str)
281 {
282 strscpy(trace_boot_options_buf, str, MAX_TRACER_SIZE);
283 return 1;
284 }
285 __setup("trace_options=", set_trace_boot_options);
286
287 static char trace_boot_clock_buf[MAX_TRACER_SIZE] __initdata;
288 static char *trace_boot_clock __initdata;
289
set_trace_boot_clock(char * str)290 static int __init set_trace_boot_clock(char *str)
291 {
292 strscpy(trace_boot_clock_buf, str, MAX_TRACER_SIZE);
293 trace_boot_clock = trace_boot_clock_buf;
294 return 1;
295 }
296 __setup("trace_clock=", set_trace_boot_clock);
297
set_tracepoint_printk(char * str)298 static int __init set_tracepoint_printk(char *str)
299 {
300 /* Ignore the "tp_printk_stop_on_boot" param */
301 if (*str == '_')
302 return 0;
303
304 if ((strcmp(str, "=0") != 0 && strcmp(str, "=off") != 0))
305 tracepoint_printk = 1;
306 return 1;
307 }
308 __setup("tp_printk", set_tracepoint_printk);
309
set_tracepoint_printk_stop(char * str)310 static int __init set_tracepoint_printk_stop(char *str)
311 {
312 tracepoint_printk_stop_on_boot = true;
313 return 1;
314 }
315 __setup("tp_printk_stop_on_boot", set_tracepoint_printk_stop);
316
ns2usecs(u64 nsec)317 unsigned long long ns2usecs(u64 nsec)
318 {
319 nsec += 500;
320 do_div(nsec, 1000);
321 return nsec;
322 }
323
324 static void
trace_process_export(struct trace_export * export,struct ring_buffer_event * event,int flag)325 trace_process_export(struct trace_export *export,
326 struct ring_buffer_event *event, int flag)
327 {
328 struct trace_entry *entry;
329 unsigned int size = 0;
330
331 if (export->flags & flag) {
332 entry = ring_buffer_event_data(event);
333 size = ring_buffer_event_length(event);
334 export->write(export, entry, size);
335 }
336 }
337
338 static DEFINE_MUTEX(ftrace_export_lock);
339
340 static struct trace_export __rcu *ftrace_exports_list __read_mostly;
341
342 static DEFINE_STATIC_KEY_FALSE(trace_function_exports_enabled);
343 static DEFINE_STATIC_KEY_FALSE(trace_event_exports_enabled);
344 static DEFINE_STATIC_KEY_FALSE(trace_marker_exports_enabled);
345
ftrace_exports_enable(struct trace_export * export)346 static inline void ftrace_exports_enable(struct trace_export *export)
347 {
348 if (export->flags & TRACE_EXPORT_FUNCTION)
349 static_branch_inc(&trace_function_exports_enabled);
350
351 if (export->flags & TRACE_EXPORT_EVENT)
352 static_branch_inc(&trace_event_exports_enabled);
353
354 if (export->flags & TRACE_EXPORT_MARKER)
355 static_branch_inc(&trace_marker_exports_enabled);
356 }
357
ftrace_exports_disable(struct trace_export * export)358 static inline void ftrace_exports_disable(struct trace_export *export)
359 {
360 if (export->flags & TRACE_EXPORT_FUNCTION)
361 static_branch_dec(&trace_function_exports_enabled);
362
363 if (export->flags & TRACE_EXPORT_EVENT)
364 static_branch_dec(&trace_event_exports_enabled);
365
366 if (export->flags & TRACE_EXPORT_MARKER)
367 static_branch_dec(&trace_marker_exports_enabled);
368 }
369
ftrace_exports(struct ring_buffer_event * event,int flag)370 static void ftrace_exports(struct ring_buffer_event *event, int flag)
371 {
372 struct trace_export *export;
373
374 preempt_disable_notrace();
375
376 export = rcu_dereference_raw_check(ftrace_exports_list);
377 while (export) {
378 trace_process_export(export, event, flag);
379 export = rcu_dereference_raw_check(export->next);
380 }
381
382 preempt_enable_notrace();
383 }
384
385 static inline void
add_trace_export(struct trace_export ** list,struct trace_export * export)386 add_trace_export(struct trace_export **list, struct trace_export *export)
387 {
388 rcu_assign_pointer(export->next, *list);
389 /*
390 * We are entering export into the list but another
391 * CPU might be walking that list. We need to make sure
392 * the export->next pointer is valid before another CPU sees
393 * the export pointer included into the list.
394 */
395 rcu_assign_pointer(*list, export);
396 }
397
398 static inline int
rm_trace_export(struct trace_export ** list,struct trace_export * export)399 rm_trace_export(struct trace_export **list, struct trace_export *export)
400 {
401 struct trace_export **p;
402
403 for (p = list; *p != NULL; p = &(*p)->next)
404 if (*p == export)
405 break;
406
407 if (*p != export)
408 return -1;
409
410 rcu_assign_pointer(*p, (*p)->next);
411
412 return 0;
413 }
414
415 static inline void
add_ftrace_export(struct trace_export ** list,struct trace_export * export)416 add_ftrace_export(struct trace_export **list, struct trace_export *export)
417 {
418 ftrace_exports_enable(export);
419
420 add_trace_export(list, export);
421 }
422
423 static inline int
rm_ftrace_export(struct trace_export ** list,struct trace_export * export)424 rm_ftrace_export(struct trace_export **list, struct trace_export *export)
425 {
426 int ret;
427
428 ret = rm_trace_export(list, export);
429 ftrace_exports_disable(export);
430
431 return ret;
432 }
433
register_ftrace_export(struct trace_export * export)434 int register_ftrace_export(struct trace_export *export)
435 {
436 if (WARN_ON_ONCE(!export->write))
437 return -1;
438
439 mutex_lock(&ftrace_export_lock);
440
441 add_ftrace_export(&ftrace_exports_list, export);
442
443 mutex_unlock(&ftrace_export_lock);
444
445 return 0;
446 }
447 EXPORT_SYMBOL_GPL(register_ftrace_export);
448
unregister_ftrace_export(struct trace_export * export)449 int unregister_ftrace_export(struct trace_export *export)
450 {
451 int ret;
452
453 mutex_lock(&ftrace_export_lock);
454
455 ret = rm_ftrace_export(&ftrace_exports_list, export);
456
457 mutex_unlock(&ftrace_export_lock);
458
459 return ret;
460 }
461 EXPORT_SYMBOL_GPL(unregister_ftrace_export);
462
463 /* trace_flags holds trace_options default values */
464 #define TRACE_DEFAULT_FLAGS \
465 (FUNCTION_DEFAULT_FLAGS | \
466 TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | \
467 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | \
468 TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | \
469 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS | \
470 TRACE_ITER_HASH_PTR)
471
472 /* trace_options that are only supported by global_trace */
473 #define TOP_LEVEL_TRACE_FLAGS (TRACE_ITER_PRINTK | \
474 TRACE_ITER_PRINTK_MSGONLY | TRACE_ITER_RECORD_CMD)
475
476 /* trace_flags that are default zero for instances */
477 #define ZEROED_TRACE_FLAGS \
478 (TRACE_ITER_EVENT_FORK | TRACE_ITER_FUNC_FORK)
479
480 /*
481 * The global_trace is the descriptor that holds the top-level tracing
482 * buffers for the live tracing.
483 */
484 static struct trace_array global_trace = {
485 .trace_flags = TRACE_DEFAULT_FLAGS,
486 };
487
trace_set_ring_buffer_expanded(struct trace_array * tr)488 void trace_set_ring_buffer_expanded(struct trace_array *tr)
489 {
490 if (!tr)
491 tr = &global_trace;
492 tr->ring_buffer_expanded = true;
493 }
494
495 LIST_HEAD(ftrace_trace_arrays);
496
trace_array_get(struct trace_array * this_tr)497 int trace_array_get(struct trace_array *this_tr)
498 {
499 struct trace_array *tr;
500 int ret = -ENODEV;
501
502 mutex_lock(&trace_types_lock);
503 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
504 if (tr == this_tr) {
505 tr->ref++;
506 ret = 0;
507 break;
508 }
509 }
510 mutex_unlock(&trace_types_lock);
511
512 return ret;
513 }
514
__trace_array_put(struct trace_array * this_tr)515 static void __trace_array_put(struct trace_array *this_tr)
516 {
517 WARN_ON(!this_tr->ref);
518 this_tr->ref--;
519 }
520
521 /**
522 * trace_array_put - Decrement the reference counter for this trace array.
523 * @this_tr : pointer to the trace array
524 *
525 * NOTE: Use this when we no longer need the trace array returned by
526 * trace_array_get_by_name(). This ensures the trace array can be later
527 * destroyed.
528 *
529 */
trace_array_put(struct trace_array * this_tr)530 void trace_array_put(struct trace_array *this_tr)
531 {
532 if (!this_tr)
533 return;
534
535 mutex_lock(&trace_types_lock);
536 __trace_array_put(this_tr);
537 mutex_unlock(&trace_types_lock);
538 }
539 EXPORT_SYMBOL_GPL(trace_array_put);
540
tracing_check_open_get_tr(struct trace_array * tr)541 int tracing_check_open_get_tr(struct trace_array *tr)
542 {
543 int ret;
544
545 ret = security_locked_down(LOCKDOWN_TRACEFS);
546 if (ret)
547 return ret;
548
549 if (tracing_disabled)
550 return -ENODEV;
551
552 if (tr && trace_array_get(tr) < 0)
553 return -ENODEV;
554
555 return 0;
556 }
557
call_filter_check_discard(struct trace_event_call * call,void * rec,struct trace_buffer * buffer,struct ring_buffer_event * event)558 int call_filter_check_discard(struct trace_event_call *call, void *rec,
559 struct trace_buffer *buffer,
560 struct ring_buffer_event *event)
561 {
562 if (unlikely(call->flags & TRACE_EVENT_FL_FILTERED) &&
563 !filter_match_preds(call->filter, rec)) {
564 __trace_event_discard_commit(buffer, event);
565 return 1;
566 }
567
568 return 0;
569 }
570
571 /**
572 * trace_find_filtered_pid - check if a pid exists in a filtered_pid list
573 * @filtered_pids: The list of pids to check
574 * @search_pid: The PID to find in @filtered_pids
575 *
576 * Returns true if @search_pid is found in @filtered_pids, and false otherwise.
577 */
578 bool
trace_find_filtered_pid(struct trace_pid_list * filtered_pids,pid_t search_pid)579 trace_find_filtered_pid(struct trace_pid_list *filtered_pids, pid_t search_pid)
580 {
581 return trace_pid_list_is_set(filtered_pids, search_pid);
582 }
583
584 /**
585 * trace_ignore_this_task - should a task be ignored for tracing
586 * @filtered_pids: The list of pids to check
587 * @filtered_no_pids: The list of pids not to be traced
588 * @task: The task that should be ignored if not filtered
589 *
590 * Checks if @task should be traced or not from @filtered_pids.
591 * Returns true if @task should *NOT* be traced.
592 * Returns false if @task should be traced.
593 */
594 bool
trace_ignore_this_task(struct trace_pid_list * filtered_pids,struct trace_pid_list * filtered_no_pids,struct task_struct * task)595 trace_ignore_this_task(struct trace_pid_list *filtered_pids,
596 struct trace_pid_list *filtered_no_pids,
597 struct task_struct *task)
598 {
599 /*
600 * If filtered_no_pids is not empty, and the task's pid is listed
601 * in filtered_no_pids, then return true.
602 * Otherwise, if filtered_pids is empty, that means we can
603 * trace all tasks. If it has content, then only trace pids
604 * within filtered_pids.
605 */
606
607 return (filtered_pids &&
608 !trace_find_filtered_pid(filtered_pids, task->pid)) ||
609 (filtered_no_pids &&
610 trace_find_filtered_pid(filtered_no_pids, task->pid));
611 }
612
613 /**
614 * trace_filter_add_remove_task - Add or remove a task from a pid_list
615 * @pid_list: The list to modify
616 * @self: The current task for fork or NULL for exit
617 * @task: The task to add or remove
618 *
619 * If adding a task, if @self is defined, the task is only added if @self
620 * is also included in @pid_list. This happens on fork and tasks should
621 * only be added when the parent is listed. If @self is NULL, then the
622 * @task pid will be removed from the list, which would happen on exit
623 * of a task.
624 */
trace_filter_add_remove_task(struct trace_pid_list * pid_list,struct task_struct * self,struct task_struct * task)625 void trace_filter_add_remove_task(struct trace_pid_list *pid_list,
626 struct task_struct *self,
627 struct task_struct *task)
628 {
629 if (!pid_list)
630 return;
631
632 /* For forks, we only add if the forking task is listed */
633 if (self) {
634 if (!trace_find_filtered_pid(pid_list, self->pid))
635 return;
636 }
637
638 /* "self" is set for forks, and NULL for exits */
639 if (self)
640 trace_pid_list_set(pid_list, task->pid);
641 else
642 trace_pid_list_clear(pid_list, task->pid);
643 }
644
645 /**
646 * trace_pid_next - Used for seq_file to get to the next pid of a pid_list
647 * @pid_list: The pid list to show
648 * @v: The last pid that was shown (+1 the actual pid to let zero be displayed)
649 * @pos: The position of the file
650 *
651 * This is used by the seq_file "next" operation to iterate the pids
652 * listed in a trace_pid_list structure.
653 *
654 * Returns the pid+1 as we want to display pid of zero, but NULL would
655 * stop the iteration.
656 */
trace_pid_next(struct trace_pid_list * pid_list,void * v,loff_t * pos)657 void *trace_pid_next(struct trace_pid_list *pid_list, void *v, loff_t *pos)
658 {
659 long pid = (unsigned long)v;
660 unsigned int next;
661
662 (*pos)++;
663
664 /* pid already is +1 of the actual previous bit */
665 if (trace_pid_list_next(pid_list, pid, &next) < 0)
666 return NULL;
667
668 pid = next;
669
670 /* Return pid + 1 to allow zero to be represented */
671 return (void *)(pid + 1);
672 }
673
674 /**
675 * trace_pid_start - Used for seq_file to start reading pid lists
676 * @pid_list: The pid list to show
677 * @pos: The position of the file
678 *
679 * This is used by seq_file "start" operation to start the iteration
680 * of listing pids.
681 *
682 * Returns the pid+1 as we want to display pid of zero, but NULL would
683 * stop the iteration.
684 */
trace_pid_start(struct trace_pid_list * pid_list,loff_t * pos)685 void *trace_pid_start(struct trace_pid_list *pid_list, loff_t *pos)
686 {
687 unsigned long pid;
688 unsigned int first;
689 loff_t l = 0;
690
691 if (trace_pid_list_first(pid_list, &first) < 0)
692 return NULL;
693
694 pid = first;
695
696 /* Return pid + 1 so that zero can be the exit value */
697 for (pid++; pid && l < *pos;
698 pid = (unsigned long)trace_pid_next(pid_list, (void *)pid, &l))
699 ;
700 return (void *)pid;
701 }
702
703 /**
704 * trace_pid_show - show the current pid in seq_file processing
705 * @m: The seq_file structure to write into
706 * @v: A void pointer of the pid (+1) value to display
707 *
708 * Can be directly used by seq_file operations to display the current
709 * pid value.
710 */
trace_pid_show(struct seq_file * m,void * v)711 int trace_pid_show(struct seq_file *m, void *v)
712 {
713 unsigned long pid = (unsigned long)v - 1;
714
715 seq_printf(m, "%lu\n", pid);
716 return 0;
717 }
718
719 /* 128 should be much more than enough */
720 #define PID_BUF_SIZE 127
721
trace_pid_write(struct trace_pid_list * filtered_pids,struct trace_pid_list ** new_pid_list,const char __user * ubuf,size_t cnt)722 int trace_pid_write(struct trace_pid_list *filtered_pids,
723 struct trace_pid_list **new_pid_list,
724 const char __user *ubuf, size_t cnt)
725 {
726 struct trace_pid_list *pid_list;
727 struct trace_parser parser;
728 unsigned long val;
729 int nr_pids = 0;
730 ssize_t read = 0;
731 ssize_t ret;
732 loff_t pos;
733 pid_t pid;
734
735 if (trace_parser_get_init(&parser, PID_BUF_SIZE + 1))
736 return -ENOMEM;
737
738 /*
739 * Always recreate a new array. The write is an all or nothing
740 * operation. Always create a new array when adding new pids by
741 * the user. If the operation fails, then the current list is
742 * not modified.
743 */
744 pid_list = trace_pid_list_alloc();
745 if (!pid_list) {
746 trace_parser_put(&parser);
747 return -ENOMEM;
748 }
749
750 if (filtered_pids) {
751 /* copy the current bits to the new max */
752 ret = trace_pid_list_first(filtered_pids, &pid);
753 while (!ret) {
754 trace_pid_list_set(pid_list, pid);
755 ret = trace_pid_list_next(filtered_pids, pid + 1, &pid);
756 nr_pids++;
757 }
758 }
759
760 ret = 0;
761 while (cnt > 0) {
762
763 pos = 0;
764
765 ret = trace_get_user(&parser, ubuf, cnt, &pos);
766 if (ret < 0)
767 break;
768
769 read += ret;
770 ubuf += ret;
771 cnt -= ret;
772
773 if (!trace_parser_loaded(&parser))
774 break;
775
776 ret = -EINVAL;
777 if (kstrtoul(parser.buffer, 0, &val))
778 break;
779
780 pid = (pid_t)val;
781
782 if (trace_pid_list_set(pid_list, pid) < 0) {
783 ret = -1;
784 break;
785 }
786 nr_pids++;
787
788 trace_parser_clear(&parser);
789 ret = 0;
790 }
791 trace_parser_put(&parser);
792
793 if (ret < 0) {
794 trace_pid_list_free(pid_list);
795 return ret;
796 }
797
798 if (!nr_pids) {
799 /* Cleared the list of pids */
800 trace_pid_list_free(pid_list);
801 pid_list = NULL;
802 }
803
804 *new_pid_list = pid_list;
805
806 return read;
807 }
808
buffer_ftrace_now(struct array_buffer * buf,int cpu)809 static u64 buffer_ftrace_now(struct array_buffer *buf, int cpu)
810 {
811 u64 ts;
812
813 /* Early boot up does not have a buffer yet */
814 if (!buf->buffer)
815 return trace_clock_local();
816
817 ts = ring_buffer_time_stamp(buf->buffer);
818 ring_buffer_normalize_time_stamp(buf->buffer, cpu, &ts);
819
820 return ts;
821 }
822
ftrace_now(int cpu)823 u64 ftrace_now(int cpu)
824 {
825 return buffer_ftrace_now(&global_trace.array_buffer, cpu);
826 }
827
828 /**
829 * tracing_is_enabled - Show if global_trace has been enabled
830 *
831 * Shows if the global trace has been enabled or not. It uses the
832 * mirror flag "buffer_disabled" to be used in fast paths such as for
833 * the irqsoff tracer. But it may be inaccurate due to races. If you
834 * need to know the accurate state, use tracing_is_on() which is a little
835 * slower, but accurate.
836 */
tracing_is_enabled(void)837 int tracing_is_enabled(void)
838 {
839 /*
840 * For quick access (irqsoff uses this in fast path), just
841 * return the mirror variable of the state of the ring buffer.
842 * It's a little racy, but we don't really care.
843 */
844 smp_rmb();
845 return !global_trace.buffer_disabled;
846 }
847
848 /*
849 * trace_buf_size is the size in bytes that is allocated
850 * for a buffer. Note, the number of bytes is always rounded
851 * to page size.
852 *
853 * This number is purposely set to a low number of 16384.
854 * If the dump on oops happens, it will be much appreciated
855 * to not have to wait for all that output. Anyway this can be
856 * boot time and run time configurable.
857 */
858 #define TRACE_BUF_SIZE_DEFAULT 1441792UL /* 16384 * 88 (sizeof(entry)) */
859
860 static unsigned long trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
861
862 /* trace_types holds a link list of available tracers. */
863 static struct tracer *trace_types __read_mostly;
864
865 /*
866 * trace_types_lock is used to protect the trace_types list.
867 */
868 DEFINE_MUTEX(trace_types_lock);
869
870 /*
871 * serialize the access of the ring buffer
872 *
873 * ring buffer serializes readers, but it is low level protection.
874 * The validity of the events (which returns by ring_buffer_peek() ..etc)
875 * are not protected by ring buffer.
876 *
877 * The content of events may become garbage if we allow other process consumes
878 * these events concurrently:
879 * A) the page of the consumed events may become a normal page
880 * (not reader page) in ring buffer, and this page will be rewritten
881 * by events producer.
882 * B) The page of the consumed events may become a page for splice_read,
883 * and this page will be returned to system.
884 *
885 * These primitives allow multi process access to different cpu ring buffer
886 * concurrently.
887 *
888 * These primitives don't distinguish read-only and read-consume access.
889 * Multi read-only access are also serialized.
890 */
891
892 #ifdef CONFIG_SMP
893 static DECLARE_RWSEM(all_cpu_access_lock);
894 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
895
trace_access_lock(int cpu)896 static inline void trace_access_lock(int cpu)
897 {
898 if (cpu == RING_BUFFER_ALL_CPUS) {
899 /* gain it for accessing the whole ring buffer. */
900 down_write(&all_cpu_access_lock);
901 } else {
902 /* gain it for accessing a cpu ring buffer. */
903
904 /* Firstly block other trace_access_lock(RING_BUFFER_ALL_CPUS). */
905 down_read(&all_cpu_access_lock);
906
907 /* Secondly block other access to this @cpu ring buffer. */
908 mutex_lock(&per_cpu(cpu_access_lock, cpu));
909 }
910 }
911
trace_access_unlock(int cpu)912 static inline void trace_access_unlock(int cpu)
913 {
914 if (cpu == RING_BUFFER_ALL_CPUS) {
915 up_write(&all_cpu_access_lock);
916 } else {
917 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
918 up_read(&all_cpu_access_lock);
919 }
920 }
921
trace_access_lock_init(void)922 static inline void trace_access_lock_init(void)
923 {
924 int cpu;
925
926 for_each_possible_cpu(cpu)
927 mutex_init(&per_cpu(cpu_access_lock, cpu));
928 }
929
930 #else
931
932 static DEFINE_MUTEX(access_lock);
933
trace_access_lock(int cpu)934 static inline void trace_access_lock(int cpu)
935 {
936 (void)cpu;
937 mutex_lock(&access_lock);
938 }
939
trace_access_unlock(int cpu)940 static inline void trace_access_unlock(int cpu)
941 {
942 (void)cpu;
943 mutex_unlock(&access_lock);
944 }
945
trace_access_lock_init(void)946 static inline void trace_access_lock_init(void)
947 {
948 }
949
950 #endif
951
952 #ifdef CONFIG_STACKTRACE
953 static void __ftrace_trace_stack(struct trace_buffer *buffer,
954 unsigned int trace_ctx,
955 int skip, struct pt_regs *regs);
956 static inline void ftrace_trace_stack(struct trace_array *tr,
957 struct trace_buffer *buffer,
958 unsigned int trace_ctx,
959 int skip, struct pt_regs *regs);
960
961 #else
__ftrace_trace_stack(struct trace_buffer * buffer,unsigned int trace_ctx,int skip,struct pt_regs * regs)962 static inline void __ftrace_trace_stack(struct trace_buffer *buffer,
963 unsigned int trace_ctx,
964 int skip, struct pt_regs *regs)
965 {
966 }
ftrace_trace_stack(struct trace_array * tr,struct trace_buffer * buffer,unsigned long trace_ctx,int skip,struct pt_regs * regs)967 static inline void ftrace_trace_stack(struct trace_array *tr,
968 struct trace_buffer *buffer,
969 unsigned long trace_ctx,
970 int skip, struct pt_regs *regs)
971 {
972 }
973
974 #endif
975
976 static __always_inline void
trace_event_setup(struct ring_buffer_event * event,int type,unsigned int trace_ctx)977 trace_event_setup(struct ring_buffer_event *event,
978 int type, unsigned int trace_ctx)
979 {
980 struct trace_entry *ent = ring_buffer_event_data(event);
981
982 tracing_generic_entry_update(ent, type, trace_ctx);
983 }
984
985 static __always_inline struct ring_buffer_event *
__trace_buffer_lock_reserve(struct trace_buffer * buffer,int type,unsigned long len,unsigned int trace_ctx)986 __trace_buffer_lock_reserve(struct trace_buffer *buffer,
987 int type,
988 unsigned long len,
989 unsigned int trace_ctx)
990 {
991 struct ring_buffer_event *event;
992
993 event = ring_buffer_lock_reserve(buffer, len);
994 if (event != NULL)
995 trace_event_setup(event, type, trace_ctx);
996
997 return event;
998 }
999
tracer_tracing_on(struct trace_array * tr)1000 void tracer_tracing_on(struct trace_array *tr)
1001 {
1002 if (tr->array_buffer.buffer)
1003 ring_buffer_record_on(tr->array_buffer.buffer);
1004 /*
1005 * This flag is looked at when buffers haven't been allocated
1006 * yet, or by some tracers (like irqsoff), that just want to
1007 * know if the ring buffer has been disabled, but it can handle
1008 * races of where it gets disabled but we still do a record.
1009 * As the check is in the fast path of the tracers, it is more
1010 * important to be fast than accurate.
1011 */
1012 tr->buffer_disabled = 0;
1013 /* Make the flag seen by readers */
1014 smp_wmb();
1015 }
1016
1017 /**
1018 * tracing_on - enable tracing buffers
1019 *
1020 * This function enables tracing buffers that may have been
1021 * disabled with tracing_off.
1022 */
tracing_on(void)1023 void tracing_on(void)
1024 {
1025 tracer_tracing_on(&global_trace);
1026 }
1027 EXPORT_SYMBOL_GPL(tracing_on);
1028
1029
1030 static __always_inline void
__buffer_unlock_commit(struct trace_buffer * buffer,struct ring_buffer_event * event)1031 __buffer_unlock_commit(struct trace_buffer *buffer, struct ring_buffer_event *event)
1032 {
1033 __this_cpu_write(trace_taskinfo_save, true);
1034
1035 /* If this is the temp buffer, we need to commit fully */
1036 if (this_cpu_read(trace_buffered_event) == event) {
1037 /* Length is in event->array[0] */
1038 ring_buffer_write(buffer, event->array[0], &event->array[1]);
1039 /* Release the temp buffer */
1040 this_cpu_dec(trace_buffered_event_cnt);
1041 /* ring_buffer_unlock_commit() enables preemption */
1042 preempt_enable_notrace();
1043 } else
1044 ring_buffer_unlock_commit(buffer);
1045 }
1046
__trace_array_puts(struct trace_array * tr,unsigned long ip,const char * str,int size)1047 int __trace_array_puts(struct trace_array *tr, unsigned long ip,
1048 const char *str, int size)
1049 {
1050 struct ring_buffer_event *event;
1051 struct trace_buffer *buffer;
1052 struct print_entry *entry;
1053 unsigned int trace_ctx;
1054 int alloc;
1055
1056 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
1057 return 0;
1058
1059 if (unlikely(tracing_selftest_running && tr == &global_trace))
1060 return 0;
1061
1062 if (unlikely(tracing_disabled))
1063 return 0;
1064
1065 alloc = sizeof(*entry) + size + 2; /* possible \n added */
1066
1067 trace_ctx = tracing_gen_ctx();
1068 buffer = tr->array_buffer.buffer;
1069 ring_buffer_nest_start(buffer);
1070 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, alloc,
1071 trace_ctx);
1072 if (!event) {
1073 size = 0;
1074 goto out;
1075 }
1076
1077 entry = ring_buffer_event_data(event);
1078 entry->ip = ip;
1079
1080 memcpy(&entry->buf, str, size);
1081
1082 /* Add a newline if necessary */
1083 if (entry->buf[size - 1] != '\n') {
1084 entry->buf[size] = '\n';
1085 entry->buf[size + 1] = '\0';
1086 } else
1087 entry->buf[size] = '\0';
1088
1089 __buffer_unlock_commit(buffer, event);
1090 ftrace_trace_stack(tr, buffer, trace_ctx, 4, NULL);
1091 out:
1092 ring_buffer_nest_end(buffer);
1093 return size;
1094 }
1095 EXPORT_SYMBOL_GPL(__trace_array_puts);
1096
1097 /**
1098 * __trace_puts - write a constant string into the trace buffer.
1099 * @ip: The address of the caller
1100 * @str: The constant string to write
1101 * @size: The size of the string.
1102 */
__trace_puts(unsigned long ip,const char * str,int size)1103 int __trace_puts(unsigned long ip, const char *str, int size)
1104 {
1105 return __trace_array_puts(&global_trace, ip, str, size);
1106 }
1107 EXPORT_SYMBOL_GPL(__trace_puts);
1108
1109 /**
1110 * __trace_bputs - write the pointer to a constant string into trace buffer
1111 * @ip: The address of the caller
1112 * @str: The constant string to write to the buffer to
1113 */
__trace_bputs(unsigned long ip,const char * str)1114 int __trace_bputs(unsigned long ip, const char *str)
1115 {
1116 struct ring_buffer_event *event;
1117 struct trace_buffer *buffer;
1118 struct bputs_entry *entry;
1119 unsigned int trace_ctx;
1120 int size = sizeof(struct bputs_entry);
1121 int ret = 0;
1122
1123 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
1124 return 0;
1125
1126 if (unlikely(tracing_selftest_running || tracing_disabled))
1127 return 0;
1128
1129 trace_ctx = tracing_gen_ctx();
1130 buffer = global_trace.array_buffer.buffer;
1131
1132 ring_buffer_nest_start(buffer);
1133 event = __trace_buffer_lock_reserve(buffer, TRACE_BPUTS, size,
1134 trace_ctx);
1135 if (!event)
1136 goto out;
1137
1138 entry = ring_buffer_event_data(event);
1139 entry->ip = ip;
1140 entry->str = str;
1141
1142 __buffer_unlock_commit(buffer, event);
1143 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 4, NULL);
1144
1145 ret = 1;
1146 out:
1147 ring_buffer_nest_end(buffer);
1148 return ret;
1149 }
1150 EXPORT_SYMBOL_GPL(__trace_bputs);
1151
1152 #ifdef CONFIG_TRACER_SNAPSHOT
tracing_snapshot_instance_cond(struct trace_array * tr,void * cond_data)1153 static void tracing_snapshot_instance_cond(struct trace_array *tr,
1154 void *cond_data)
1155 {
1156 struct tracer *tracer = tr->current_trace;
1157 unsigned long flags;
1158
1159 if (in_nmi()) {
1160 trace_array_puts(tr, "*** SNAPSHOT CALLED FROM NMI CONTEXT ***\n");
1161 trace_array_puts(tr, "*** snapshot is being ignored ***\n");
1162 return;
1163 }
1164
1165 if (!tr->allocated_snapshot) {
1166 trace_array_puts(tr, "*** SNAPSHOT NOT ALLOCATED ***\n");
1167 trace_array_puts(tr, "*** stopping trace here! ***\n");
1168 tracer_tracing_off(tr);
1169 return;
1170 }
1171
1172 /* Note, snapshot can not be used when the tracer uses it */
1173 if (tracer->use_max_tr) {
1174 trace_array_puts(tr, "*** LATENCY TRACER ACTIVE ***\n");
1175 trace_array_puts(tr, "*** Can not use snapshot (sorry) ***\n");
1176 return;
1177 }
1178
1179 local_irq_save(flags);
1180 update_max_tr(tr, current, smp_processor_id(), cond_data);
1181 local_irq_restore(flags);
1182 }
1183
tracing_snapshot_instance(struct trace_array * tr)1184 void tracing_snapshot_instance(struct trace_array *tr)
1185 {
1186 tracing_snapshot_instance_cond(tr, NULL);
1187 }
1188
1189 /**
1190 * tracing_snapshot - take a snapshot of the current buffer.
1191 *
1192 * This causes a swap between the snapshot buffer and the current live
1193 * tracing buffer. You can use this to take snapshots of the live
1194 * trace when some condition is triggered, but continue to trace.
1195 *
1196 * Note, make sure to allocate the snapshot with either
1197 * a tracing_snapshot_alloc(), or by doing it manually
1198 * with: echo 1 > /sys/kernel/tracing/snapshot
1199 *
1200 * If the snapshot buffer is not allocated, it will stop tracing.
1201 * Basically making a permanent snapshot.
1202 */
tracing_snapshot(void)1203 void tracing_snapshot(void)
1204 {
1205 struct trace_array *tr = &global_trace;
1206
1207 tracing_snapshot_instance(tr);
1208 }
1209 EXPORT_SYMBOL_GPL(tracing_snapshot);
1210
1211 /**
1212 * tracing_snapshot_cond - conditionally take a snapshot of the current buffer.
1213 * @tr: The tracing instance to snapshot
1214 * @cond_data: The data to be tested conditionally, and possibly saved
1215 *
1216 * This is the same as tracing_snapshot() except that the snapshot is
1217 * conditional - the snapshot will only happen if the
1218 * cond_snapshot.update() implementation receiving the cond_data
1219 * returns true, which means that the trace array's cond_snapshot
1220 * update() operation used the cond_data to determine whether the
1221 * snapshot should be taken, and if it was, presumably saved it along
1222 * with the snapshot.
1223 */
tracing_snapshot_cond(struct trace_array * tr,void * cond_data)1224 void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1225 {
1226 tracing_snapshot_instance_cond(tr, cond_data);
1227 }
1228 EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
1229
1230 /**
1231 * tracing_cond_snapshot_data - get the user data associated with a snapshot
1232 * @tr: The tracing instance
1233 *
1234 * When the user enables a conditional snapshot using
1235 * tracing_snapshot_cond_enable(), the user-defined cond_data is saved
1236 * with the snapshot. This accessor is used to retrieve it.
1237 *
1238 * Should not be called from cond_snapshot.update(), since it takes
1239 * the tr->max_lock lock, which the code calling
1240 * cond_snapshot.update() has already done.
1241 *
1242 * Returns the cond_data associated with the trace array's snapshot.
1243 */
tracing_cond_snapshot_data(struct trace_array * tr)1244 void *tracing_cond_snapshot_data(struct trace_array *tr)
1245 {
1246 void *cond_data = NULL;
1247
1248 local_irq_disable();
1249 arch_spin_lock(&tr->max_lock);
1250
1251 if (tr->cond_snapshot)
1252 cond_data = tr->cond_snapshot->cond_data;
1253
1254 arch_spin_unlock(&tr->max_lock);
1255 local_irq_enable();
1256
1257 return cond_data;
1258 }
1259 EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
1260
1261 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
1262 struct array_buffer *size_buf, int cpu_id);
1263 static void set_buffer_entries(struct array_buffer *buf, unsigned long val);
1264
tracing_alloc_snapshot_instance(struct trace_array * tr)1265 int tracing_alloc_snapshot_instance(struct trace_array *tr)
1266 {
1267 int order;
1268 int ret;
1269
1270 if (!tr->allocated_snapshot) {
1271
1272 /* Make the snapshot buffer have the same order as main buffer */
1273 order = ring_buffer_subbuf_order_get(tr->array_buffer.buffer);
1274 ret = ring_buffer_subbuf_order_set(tr->max_buffer.buffer, order);
1275 if (ret < 0)
1276 return ret;
1277
1278 /* allocate spare buffer */
1279 ret = resize_buffer_duplicate_size(&tr->max_buffer,
1280 &tr->array_buffer, RING_BUFFER_ALL_CPUS);
1281 if (ret < 0)
1282 return ret;
1283
1284 tr->allocated_snapshot = true;
1285 }
1286
1287 return 0;
1288 }
1289
free_snapshot(struct trace_array * tr)1290 static void free_snapshot(struct trace_array *tr)
1291 {
1292 /*
1293 * We don't free the ring buffer. instead, resize it because
1294 * The max_tr ring buffer has some state (e.g. ring->clock) and
1295 * we want preserve it.
1296 */
1297 ring_buffer_subbuf_order_set(tr->max_buffer.buffer, 0);
1298 ring_buffer_resize(tr->max_buffer.buffer, 1, RING_BUFFER_ALL_CPUS);
1299 set_buffer_entries(&tr->max_buffer, 1);
1300 tracing_reset_online_cpus(&tr->max_buffer);
1301 tr->allocated_snapshot = false;
1302 }
1303
1304 /**
1305 * tracing_alloc_snapshot - allocate snapshot buffer.
1306 *
1307 * This only allocates the snapshot buffer if it isn't already
1308 * allocated - it doesn't also take a snapshot.
1309 *
1310 * This is meant to be used in cases where the snapshot buffer needs
1311 * to be set up for events that can't sleep but need to be able to
1312 * trigger a snapshot.
1313 */
tracing_alloc_snapshot(void)1314 int tracing_alloc_snapshot(void)
1315 {
1316 struct trace_array *tr = &global_trace;
1317 int ret;
1318
1319 ret = tracing_alloc_snapshot_instance(tr);
1320 WARN_ON(ret < 0);
1321
1322 return ret;
1323 }
1324 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
1325
1326 /**
1327 * tracing_snapshot_alloc - allocate and take a snapshot of the current buffer.
1328 *
1329 * This is similar to tracing_snapshot(), but it will allocate the
1330 * snapshot buffer if it isn't already allocated. Use this only
1331 * where it is safe to sleep, as the allocation may sleep.
1332 *
1333 * This causes a swap between the snapshot buffer and the current live
1334 * tracing buffer. You can use this to take snapshots of the live
1335 * trace when some condition is triggered, but continue to trace.
1336 */
tracing_snapshot_alloc(void)1337 void tracing_snapshot_alloc(void)
1338 {
1339 int ret;
1340
1341 ret = tracing_alloc_snapshot();
1342 if (ret < 0)
1343 return;
1344
1345 tracing_snapshot();
1346 }
1347 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
1348
1349 /**
1350 * tracing_snapshot_cond_enable - enable conditional snapshot for an instance
1351 * @tr: The tracing instance
1352 * @cond_data: User data to associate with the snapshot
1353 * @update: Implementation of the cond_snapshot update function
1354 *
1355 * Check whether the conditional snapshot for the given instance has
1356 * already been enabled, or if the current tracer is already using a
1357 * snapshot; if so, return -EBUSY, else create a cond_snapshot and
1358 * save the cond_data and update function inside.
1359 *
1360 * Returns 0 if successful, error otherwise.
1361 */
tracing_snapshot_cond_enable(struct trace_array * tr,void * cond_data,cond_update_fn_t update)1362 int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data,
1363 cond_update_fn_t update)
1364 {
1365 struct cond_snapshot *cond_snapshot;
1366 int ret = 0;
1367
1368 cond_snapshot = kzalloc(sizeof(*cond_snapshot), GFP_KERNEL);
1369 if (!cond_snapshot)
1370 return -ENOMEM;
1371
1372 cond_snapshot->cond_data = cond_data;
1373 cond_snapshot->update = update;
1374
1375 mutex_lock(&trace_types_lock);
1376
1377 ret = tracing_alloc_snapshot_instance(tr);
1378 if (ret)
1379 goto fail_unlock;
1380
1381 if (tr->current_trace->use_max_tr) {
1382 ret = -EBUSY;
1383 goto fail_unlock;
1384 }
1385
1386 /*
1387 * The cond_snapshot can only change to NULL without the
1388 * trace_types_lock. We don't care if we race with it going
1389 * to NULL, but we want to make sure that it's not set to
1390 * something other than NULL when we get here, which we can
1391 * do safely with only holding the trace_types_lock and not
1392 * having to take the max_lock.
1393 */
1394 if (tr->cond_snapshot) {
1395 ret = -EBUSY;
1396 goto fail_unlock;
1397 }
1398
1399 local_irq_disable();
1400 arch_spin_lock(&tr->max_lock);
1401 tr->cond_snapshot = cond_snapshot;
1402 arch_spin_unlock(&tr->max_lock);
1403 local_irq_enable();
1404
1405 mutex_unlock(&trace_types_lock);
1406
1407 return ret;
1408
1409 fail_unlock:
1410 mutex_unlock(&trace_types_lock);
1411 kfree(cond_snapshot);
1412 return ret;
1413 }
1414 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
1415
1416 /**
1417 * tracing_snapshot_cond_disable - disable conditional snapshot for an instance
1418 * @tr: The tracing instance
1419 *
1420 * Check whether the conditional snapshot for the given instance is
1421 * enabled; if so, free the cond_snapshot associated with it,
1422 * otherwise return -EINVAL.
1423 *
1424 * Returns 0 if successful, error otherwise.
1425 */
tracing_snapshot_cond_disable(struct trace_array * tr)1426 int tracing_snapshot_cond_disable(struct trace_array *tr)
1427 {
1428 int ret = 0;
1429
1430 local_irq_disable();
1431 arch_spin_lock(&tr->max_lock);
1432
1433 if (!tr->cond_snapshot)
1434 ret = -EINVAL;
1435 else {
1436 kfree(tr->cond_snapshot);
1437 tr->cond_snapshot = NULL;
1438 }
1439
1440 arch_spin_unlock(&tr->max_lock);
1441 local_irq_enable();
1442
1443 return ret;
1444 }
1445 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1446 #else
tracing_snapshot(void)1447 void tracing_snapshot(void)
1448 {
1449 WARN_ONCE(1, "Snapshot feature not enabled, but internal snapshot used");
1450 }
1451 EXPORT_SYMBOL_GPL(tracing_snapshot);
tracing_snapshot_cond(struct trace_array * tr,void * cond_data)1452 void tracing_snapshot_cond(struct trace_array *tr, void *cond_data)
1453 {
1454 WARN_ONCE(1, "Snapshot feature not enabled, but internal conditional snapshot used");
1455 }
1456 EXPORT_SYMBOL_GPL(tracing_snapshot_cond);
tracing_alloc_snapshot(void)1457 int tracing_alloc_snapshot(void)
1458 {
1459 WARN_ONCE(1, "Snapshot feature not enabled, but snapshot allocation used");
1460 return -ENODEV;
1461 }
1462 EXPORT_SYMBOL_GPL(tracing_alloc_snapshot);
tracing_snapshot_alloc(void)1463 void tracing_snapshot_alloc(void)
1464 {
1465 /* Give warning */
1466 tracing_snapshot();
1467 }
1468 EXPORT_SYMBOL_GPL(tracing_snapshot_alloc);
tracing_cond_snapshot_data(struct trace_array * tr)1469 void *tracing_cond_snapshot_data(struct trace_array *tr)
1470 {
1471 return NULL;
1472 }
1473 EXPORT_SYMBOL_GPL(tracing_cond_snapshot_data);
tracing_snapshot_cond_enable(struct trace_array * tr,void * cond_data,cond_update_fn_t update)1474 int tracing_snapshot_cond_enable(struct trace_array *tr, void *cond_data, cond_update_fn_t update)
1475 {
1476 return -ENODEV;
1477 }
1478 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_enable);
tracing_snapshot_cond_disable(struct trace_array * tr)1479 int tracing_snapshot_cond_disable(struct trace_array *tr)
1480 {
1481 return false;
1482 }
1483 EXPORT_SYMBOL_GPL(tracing_snapshot_cond_disable);
1484 #define free_snapshot(tr) do { } while (0)
1485 #endif /* CONFIG_TRACER_SNAPSHOT */
1486
tracer_tracing_off(struct trace_array * tr)1487 void tracer_tracing_off(struct trace_array *tr)
1488 {
1489 if (tr->array_buffer.buffer)
1490 ring_buffer_record_off(tr->array_buffer.buffer);
1491 /*
1492 * This flag is looked at when buffers haven't been allocated
1493 * yet, or by some tracers (like irqsoff), that just want to
1494 * know if the ring buffer has been disabled, but it can handle
1495 * races of where it gets disabled but we still do a record.
1496 * As the check is in the fast path of the tracers, it is more
1497 * important to be fast than accurate.
1498 */
1499 tr->buffer_disabled = 1;
1500 /* Make the flag seen by readers */
1501 smp_wmb();
1502 }
1503
1504 /**
1505 * tracing_off - turn off tracing buffers
1506 *
1507 * This function stops the tracing buffers from recording data.
1508 * It does not disable any overhead the tracers themselves may
1509 * be causing. This function simply causes all recording to
1510 * the ring buffers to fail.
1511 */
tracing_off(void)1512 void tracing_off(void)
1513 {
1514 tracer_tracing_off(&global_trace);
1515 }
1516 EXPORT_SYMBOL_GPL(tracing_off);
1517
disable_trace_on_warning(void)1518 void disable_trace_on_warning(void)
1519 {
1520 if (__disable_trace_on_warning) {
1521 trace_array_printk_buf(global_trace.array_buffer.buffer, _THIS_IP_,
1522 "Disabling tracing due to warning\n");
1523 tracing_off();
1524 }
1525 }
1526
1527 /**
1528 * tracer_tracing_is_on - show real state of ring buffer enabled
1529 * @tr : the trace array to know if ring buffer is enabled
1530 *
1531 * Shows real state of the ring buffer if it is enabled or not.
1532 */
tracer_tracing_is_on(struct trace_array * tr)1533 bool tracer_tracing_is_on(struct trace_array *tr)
1534 {
1535 if (tr->array_buffer.buffer)
1536 return ring_buffer_record_is_set_on(tr->array_buffer.buffer);
1537 return !tr->buffer_disabled;
1538 }
1539
1540 /**
1541 * tracing_is_on - show state of ring buffers enabled
1542 */
tracing_is_on(void)1543 int tracing_is_on(void)
1544 {
1545 return tracer_tracing_is_on(&global_trace);
1546 }
1547 EXPORT_SYMBOL_GPL(tracing_is_on);
1548
set_buf_size(char * str)1549 static int __init set_buf_size(char *str)
1550 {
1551 unsigned long buf_size;
1552
1553 if (!str)
1554 return 0;
1555 buf_size = memparse(str, &str);
1556 /*
1557 * nr_entries can not be zero and the startup
1558 * tests require some buffer space. Therefore
1559 * ensure we have at least 4096 bytes of buffer.
1560 */
1561 trace_buf_size = max(4096UL, buf_size);
1562 return 1;
1563 }
1564 __setup("trace_buf_size=", set_buf_size);
1565
set_tracing_thresh(char * str)1566 static int __init set_tracing_thresh(char *str)
1567 {
1568 unsigned long threshold;
1569 int ret;
1570
1571 if (!str)
1572 return 0;
1573 ret = kstrtoul(str, 0, &threshold);
1574 if (ret < 0)
1575 return 0;
1576 tracing_thresh = threshold * 1000;
1577 return 1;
1578 }
1579 __setup("tracing_thresh=", set_tracing_thresh);
1580
nsecs_to_usecs(unsigned long nsecs)1581 unsigned long nsecs_to_usecs(unsigned long nsecs)
1582 {
1583 return nsecs / 1000;
1584 }
1585
1586 /*
1587 * TRACE_FLAGS is defined as a tuple matching bit masks with strings.
1588 * It uses C(a, b) where 'a' is the eval (enum) name and 'b' is the string that
1589 * matches it. By defining "C(a, b) b", TRACE_FLAGS becomes a list
1590 * of strings in the order that the evals (enum) were defined.
1591 */
1592 #undef C
1593 #define C(a, b) b
1594
1595 /* These must match the bit positions in trace_iterator_flags */
1596 static const char *trace_options[] = {
1597 TRACE_FLAGS
1598 NULL
1599 };
1600
1601 static struct {
1602 u64 (*func)(void);
1603 const char *name;
1604 int in_ns; /* is this clock in nanoseconds? */
1605 } trace_clocks[] = {
1606 { trace_clock_local, "local", 1 },
1607 { trace_clock_global, "global", 1 },
1608 { trace_clock_counter, "counter", 0 },
1609 { trace_clock_jiffies, "uptime", 0 },
1610 { trace_clock, "perf", 1 },
1611 { ktime_get_mono_fast_ns, "mono", 1 },
1612 { ktime_get_raw_fast_ns, "mono_raw", 1 },
1613 { ktime_get_boot_fast_ns, "boot", 1 },
1614 { ktime_get_tai_fast_ns, "tai", 1 },
1615 ARCH_TRACE_CLOCKS
1616 };
1617
trace_clock_in_ns(struct trace_array * tr)1618 bool trace_clock_in_ns(struct trace_array *tr)
1619 {
1620 if (trace_clocks[tr->clock_id].in_ns)
1621 return true;
1622
1623 return false;
1624 }
1625
1626 /*
1627 * trace_parser_get_init - gets the buffer for trace parser
1628 */
trace_parser_get_init(struct trace_parser * parser,int size)1629 int trace_parser_get_init(struct trace_parser *parser, int size)
1630 {
1631 memset(parser, 0, sizeof(*parser));
1632
1633 parser->buffer = kmalloc(size, GFP_KERNEL);
1634 if (!parser->buffer)
1635 return 1;
1636
1637 parser->size = size;
1638 return 0;
1639 }
1640
1641 /*
1642 * trace_parser_put - frees the buffer for trace parser
1643 */
trace_parser_put(struct trace_parser * parser)1644 void trace_parser_put(struct trace_parser *parser)
1645 {
1646 kfree(parser->buffer);
1647 parser->buffer = NULL;
1648 }
1649
1650 /*
1651 * trace_get_user - reads the user input string separated by space
1652 * (matched by isspace(ch))
1653 *
1654 * For each string found the 'struct trace_parser' is updated,
1655 * and the function returns.
1656 *
1657 * Returns number of bytes read.
1658 *
1659 * See kernel/trace/trace.h for 'struct trace_parser' details.
1660 */
trace_get_user(struct trace_parser * parser,const char __user * ubuf,size_t cnt,loff_t * ppos)1661 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
1662 size_t cnt, loff_t *ppos)
1663 {
1664 char ch;
1665 size_t read = 0;
1666 ssize_t ret;
1667
1668 if (!*ppos)
1669 trace_parser_clear(parser);
1670
1671 ret = get_user(ch, ubuf++);
1672 if (ret)
1673 goto out;
1674
1675 read++;
1676 cnt--;
1677
1678 /*
1679 * The parser is not finished with the last write,
1680 * continue reading the user input without skipping spaces.
1681 */
1682 if (!parser->cont) {
1683 /* skip white space */
1684 while (cnt && isspace(ch)) {
1685 ret = get_user(ch, ubuf++);
1686 if (ret)
1687 goto out;
1688 read++;
1689 cnt--;
1690 }
1691
1692 parser->idx = 0;
1693
1694 /* only spaces were written */
1695 if (isspace(ch) || !ch) {
1696 *ppos += read;
1697 ret = read;
1698 goto out;
1699 }
1700 }
1701
1702 /* read the non-space input */
1703 while (cnt && !isspace(ch) && ch) {
1704 if (parser->idx < parser->size - 1)
1705 parser->buffer[parser->idx++] = ch;
1706 else {
1707 ret = -EINVAL;
1708 goto out;
1709 }
1710 ret = get_user(ch, ubuf++);
1711 if (ret)
1712 goto out;
1713 read++;
1714 cnt--;
1715 }
1716
1717 /* We either got finished input or we have to wait for another call. */
1718 if (isspace(ch) || !ch) {
1719 parser->buffer[parser->idx] = 0;
1720 parser->cont = false;
1721 } else if (parser->idx < parser->size - 1) {
1722 parser->cont = true;
1723 parser->buffer[parser->idx++] = ch;
1724 /* Make sure the parsed string always terminates with '\0'. */
1725 parser->buffer[parser->idx] = 0;
1726 } else {
1727 ret = -EINVAL;
1728 goto out;
1729 }
1730
1731 *ppos += read;
1732 ret = read;
1733
1734 out:
1735 return ret;
1736 }
1737
1738 /* TODO add a seq_buf_to_buffer() */
trace_seq_to_buffer(struct trace_seq * s,void * buf,size_t cnt)1739 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
1740 {
1741 int len;
1742
1743 if (trace_seq_used(s) <= s->readpos)
1744 return -EBUSY;
1745
1746 len = trace_seq_used(s) - s->readpos;
1747 if (cnt > len)
1748 cnt = len;
1749 memcpy(buf, s->buffer + s->readpos, cnt);
1750
1751 s->readpos += cnt;
1752 return cnt;
1753 }
1754
1755 unsigned long __read_mostly tracing_thresh;
1756
1757 #ifdef CONFIG_TRACER_MAX_TRACE
1758 static const struct file_operations tracing_max_lat_fops;
1759
1760 #ifdef LATENCY_FS_NOTIFY
1761
1762 static struct workqueue_struct *fsnotify_wq;
1763
latency_fsnotify_workfn(struct work_struct * work)1764 static void latency_fsnotify_workfn(struct work_struct *work)
1765 {
1766 struct trace_array *tr = container_of(work, struct trace_array,
1767 fsnotify_work);
1768 fsnotify_inode(tr->d_max_latency->d_inode, FS_MODIFY);
1769 }
1770
latency_fsnotify_workfn_irq(struct irq_work * iwork)1771 static void latency_fsnotify_workfn_irq(struct irq_work *iwork)
1772 {
1773 struct trace_array *tr = container_of(iwork, struct trace_array,
1774 fsnotify_irqwork);
1775 queue_work(fsnotify_wq, &tr->fsnotify_work);
1776 }
1777
trace_create_maxlat_file(struct trace_array * tr,struct dentry * d_tracer)1778 static void trace_create_maxlat_file(struct trace_array *tr,
1779 struct dentry *d_tracer)
1780 {
1781 INIT_WORK(&tr->fsnotify_work, latency_fsnotify_workfn);
1782 init_irq_work(&tr->fsnotify_irqwork, latency_fsnotify_workfn_irq);
1783 tr->d_max_latency = trace_create_file("tracing_max_latency",
1784 TRACE_MODE_WRITE,
1785 d_tracer, tr,
1786 &tracing_max_lat_fops);
1787 }
1788
latency_fsnotify_init(void)1789 __init static int latency_fsnotify_init(void)
1790 {
1791 fsnotify_wq = alloc_workqueue("tr_max_lat_wq",
1792 WQ_UNBOUND | WQ_HIGHPRI, 0);
1793 if (!fsnotify_wq) {
1794 pr_err("Unable to allocate tr_max_lat_wq\n");
1795 return -ENOMEM;
1796 }
1797 return 0;
1798 }
1799
1800 late_initcall_sync(latency_fsnotify_init);
1801
latency_fsnotify(struct trace_array * tr)1802 void latency_fsnotify(struct trace_array *tr)
1803 {
1804 if (!fsnotify_wq)
1805 return;
1806 /*
1807 * We cannot call queue_work(&tr->fsnotify_work) from here because it's
1808 * possible that we are called from __schedule() or do_idle(), which
1809 * could cause a deadlock.
1810 */
1811 irq_work_queue(&tr->fsnotify_irqwork);
1812 }
1813
1814 #else /* !LATENCY_FS_NOTIFY */
1815
1816 #define trace_create_maxlat_file(tr, d_tracer) \
1817 trace_create_file("tracing_max_latency", TRACE_MODE_WRITE, \
1818 d_tracer, tr, &tracing_max_lat_fops)
1819
1820 #endif
1821
1822 /*
1823 * Copy the new maximum trace into the separate maximum-trace
1824 * structure. (this way the maximum trace is permanently saved,
1825 * for later retrieval via /sys/kernel/tracing/tracing_max_latency)
1826 */
1827 static void
__update_max_tr(struct trace_array * tr,struct task_struct * tsk,int cpu)1828 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
1829 {
1830 struct array_buffer *trace_buf = &tr->array_buffer;
1831 struct array_buffer *max_buf = &tr->max_buffer;
1832 struct trace_array_cpu *data = per_cpu_ptr(trace_buf->data, cpu);
1833 struct trace_array_cpu *max_data = per_cpu_ptr(max_buf->data, cpu);
1834
1835 max_buf->cpu = cpu;
1836 max_buf->time_start = data->preempt_timestamp;
1837
1838 max_data->saved_latency = tr->max_latency;
1839 max_data->critical_start = data->critical_start;
1840 max_data->critical_end = data->critical_end;
1841
1842 strncpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
1843 max_data->pid = tsk->pid;
1844 /*
1845 * If tsk == current, then use current_uid(), as that does not use
1846 * RCU. The irq tracer can be called out of RCU scope.
1847 */
1848 if (tsk == current)
1849 max_data->uid = current_uid();
1850 else
1851 max_data->uid = task_uid(tsk);
1852
1853 max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
1854 max_data->policy = tsk->policy;
1855 max_data->rt_priority = tsk->rt_priority;
1856
1857 /* record this tasks comm */
1858 tracing_record_cmdline(tsk);
1859 latency_fsnotify(tr);
1860 }
1861
1862 /**
1863 * update_max_tr - snapshot all trace buffers from global_trace to max_tr
1864 * @tr: tracer
1865 * @tsk: the task with the latency
1866 * @cpu: The cpu that initiated the trace.
1867 * @cond_data: User data associated with a conditional snapshot
1868 *
1869 * Flip the buffers between the @tr and the max_tr and record information
1870 * about which task was the cause of this latency.
1871 */
1872 void
update_max_tr(struct trace_array * tr,struct task_struct * tsk,int cpu,void * cond_data)1873 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu,
1874 void *cond_data)
1875 {
1876 if (tr->stop_count)
1877 return;
1878
1879 WARN_ON_ONCE(!irqs_disabled());
1880
1881 if (!tr->allocated_snapshot) {
1882 /* Only the nop tracer should hit this when disabling */
1883 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1884 return;
1885 }
1886
1887 arch_spin_lock(&tr->max_lock);
1888
1889 /* Inherit the recordable setting from array_buffer */
1890 if (ring_buffer_record_is_set_on(tr->array_buffer.buffer))
1891 ring_buffer_record_on(tr->max_buffer.buffer);
1892 else
1893 ring_buffer_record_off(tr->max_buffer.buffer);
1894
1895 #ifdef CONFIG_TRACER_SNAPSHOT
1896 if (tr->cond_snapshot && !tr->cond_snapshot->update(tr, cond_data)) {
1897 arch_spin_unlock(&tr->max_lock);
1898 return;
1899 }
1900 #endif
1901 swap(tr->array_buffer.buffer, tr->max_buffer.buffer);
1902
1903 __update_max_tr(tr, tsk, cpu);
1904
1905 arch_spin_unlock(&tr->max_lock);
1906
1907 /* Any waiters on the old snapshot buffer need to wake up */
1908 ring_buffer_wake_waiters(tr->array_buffer.buffer, RING_BUFFER_ALL_CPUS);
1909 }
1910
1911 /**
1912 * update_max_tr_single - only copy one trace over, and reset the rest
1913 * @tr: tracer
1914 * @tsk: task with the latency
1915 * @cpu: the cpu of the buffer to copy.
1916 *
1917 * Flip the trace of a single CPU buffer between the @tr and the max_tr.
1918 */
1919 void
update_max_tr_single(struct trace_array * tr,struct task_struct * tsk,int cpu)1920 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
1921 {
1922 int ret;
1923
1924 if (tr->stop_count)
1925 return;
1926
1927 WARN_ON_ONCE(!irqs_disabled());
1928 if (!tr->allocated_snapshot) {
1929 /* Only the nop tracer should hit this when disabling */
1930 WARN_ON_ONCE(tr->current_trace != &nop_trace);
1931 return;
1932 }
1933
1934 arch_spin_lock(&tr->max_lock);
1935
1936 ret = ring_buffer_swap_cpu(tr->max_buffer.buffer, tr->array_buffer.buffer, cpu);
1937
1938 if (ret == -EBUSY) {
1939 /*
1940 * We failed to swap the buffer due to a commit taking
1941 * place on this CPU. We fail to record, but we reset
1942 * the max trace buffer (no one writes directly to it)
1943 * and flag that it failed.
1944 * Another reason is resize is in progress.
1945 */
1946 trace_array_printk_buf(tr->max_buffer.buffer, _THIS_IP_,
1947 "Failed to swap buffers due to commit or resize in progress\n");
1948 }
1949
1950 WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
1951
1952 __update_max_tr(tr, tsk, cpu);
1953 arch_spin_unlock(&tr->max_lock);
1954 }
1955
1956 #endif /* CONFIG_TRACER_MAX_TRACE */
1957
wait_on_pipe(struct trace_iterator * iter,int full)1958 static int wait_on_pipe(struct trace_iterator *iter, int full)
1959 {
1960 int ret;
1961
1962 /* Iterators are static, they should be filled or empty */
1963 if (trace_buffer_iter(iter, iter->cpu_file))
1964 return 0;
1965
1966 ret = ring_buffer_wait(iter->array_buffer->buffer, iter->cpu_file, full);
1967
1968 #ifdef CONFIG_TRACER_MAX_TRACE
1969 /*
1970 * Make sure this is still the snapshot buffer, as if a snapshot were
1971 * to happen, this would now be the main buffer.
1972 */
1973 if (iter->snapshot)
1974 iter->array_buffer = &iter->tr->max_buffer;
1975 #endif
1976 return ret;
1977 }
1978
1979 #ifdef CONFIG_FTRACE_STARTUP_TEST
1980 static bool selftests_can_run;
1981
1982 struct trace_selftests {
1983 struct list_head list;
1984 struct tracer *type;
1985 };
1986
1987 static LIST_HEAD(postponed_selftests);
1988
save_selftest(struct tracer * type)1989 static int save_selftest(struct tracer *type)
1990 {
1991 struct trace_selftests *selftest;
1992
1993 selftest = kmalloc(sizeof(*selftest), GFP_KERNEL);
1994 if (!selftest)
1995 return -ENOMEM;
1996
1997 selftest->type = type;
1998 list_add(&selftest->list, &postponed_selftests);
1999 return 0;
2000 }
2001
run_tracer_selftest(struct tracer * type)2002 static int run_tracer_selftest(struct tracer *type)
2003 {
2004 struct trace_array *tr = &global_trace;
2005 struct tracer *saved_tracer = tr->current_trace;
2006 int ret;
2007
2008 if (!type->selftest || tracing_selftest_disabled)
2009 return 0;
2010
2011 /*
2012 * If a tracer registers early in boot up (before scheduling is
2013 * initialized and such), then do not run its selftests yet.
2014 * Instead, run it a little later in the boot process.
2015 */
2016 if (!selftests_can_run)
2017 return save_selftest(type);
2018
2019 if (!tracing_is_on()) {
2020 pr_warn("Selftest for tracer %s skipped due to tracing disabled\n",
2021 type->name);
2022 return 0;
2023 }
2024
2025 /*
2026 * Run a selftest on this tracer.
2027 * Here we reset the trace buffer, and set the current
2028 * tracer to be this tracer. The tracer can then run some
2029 * internal tracing to verify that everything is in order.
2030 * If we fail, we do not register this tracer.
2031 */
2032 tracing_reset_online_cpus(&tr->array_buffer);
2033
2034 tr->current_trace = type;
2035
2036 #ifdef CONFIG_TRACER_MAX_TRACE
2037 if (type->use_max_tr) {
2038 /* If we expanded the buffers, make sure the max is expanded too */
2039 if (tr->ring_buffer_expanded)
2040 ring_buffer_resize(tr->max_buffer.buffer, trace_buf_size,
2041 RING_BUFFER_ALL_CPUS);
2042 tr->allocated_snapshot = true;
2043 }
2044 #endif
2045
2046 /* the test is responsible for initializing and enabling */
2047 pr_info("Testing tracer %s: ", type->name);
2048 ret = type->selftest(type, tr);
2049 /* the test is responsible for resetting too */
2050 tr->current_trace = saved_tracer;
2051 if (ret) {
2052 printk(KERN_CONT "FAILED!\n");
2053 /* Add the warning after printing 'FAILED' */
2054 WARN_ON(1);
2055 return -1;
2056 }
2057 /* Only reset on passing, to avoid touching corrupted buffers */
2058 tracing_reset_online_cpus(&tr->array_buffer);
2059
2060 #ifdef CONFIG_TRACER_MAX_TRACE
2061 if (type->use_max_tr) {
2062 tr->allocated_snapshot = false;
2063
2064 /* Shrink the max buffer again */
2065 if (tr->ring_buffer_expanded)
2066 ring_buffer_resize(tr->max_buffer.buffer, 1,
2067 RING_BUFFER_ALL_CPUS);
2068 }
2069 #endif
2070
2071 printk(KERN_CONT "PASSED\n");
2072 return 0;
2073 }
2074
do_run_tracer_selftest(struct tracer * type)2075 static int do_run_tracer_selftest(struct tracer *type)
2076 {
2077 int ret;
2078
2079 /*
2080 * Tests can take a long time, especially if they are run one after the
2081 * other, as does happen during bootup when all the tracers are
2082 * registered. This could cause the soft lockup watchdog to trigger.
2083 */
2084 cond_resched();
2085
2086 tracing_selftest_running = true;
2087 ret = run_tracer_selftest(type);
2088 tracing_selftest_running = false;
2089
2090 return ret;
2091 }
2092
init_trace_selftests(void)2093 static __init int init_trace_selftests(void)
2094 {
2095 struct trace_selftests *p, *n;
2096 struct tracer *t, **last;
2097 int ret;
2098
2099 selftests_can_run = true;
2100
2101 mutex_lock(&trace_types_lock);
2102
2103 if (list_empty(&postponed_selftests))
2104 goto out;
2105
2106 pr_info("Running postponed tracer tests:\n");
2107
2108 tracing_selftest_running = true;
2109 list_for_each_entry_safe(p, n, &postponed_selftests, list) {
2110 /* This loop can take minutes when sanitizers are enabled, so
2111 * lets make sure we allow RCU processing.
2112 */
2113 cond_resched();
2114 ret = run_tracer_selftest(p->type);
2115 /* If the test fails, then warn and remove from available_tracers */
2116 if (ret < 0) {
2117 WARN(1, "tracer: %s failed selftest, disabling\n",
2118 p->type->name);
2119 last = &trace_types;
2120 for (t = trace_types; t; t = t->next) {
2121 if (t == p->type) {
2122 *last = t->next;
2123 break;
2124 }
2125 last = &t->next;
2126 }
2127 }
2128 list_del(&p->list);
2129 kfree(p);
2130 }
2131 tracing_selftest_running = false;
2132
2133 out:
2134 mutex_unlock(&trace_types_lock);
2135
2136 return 0;
2137 }
2138 core_initcall(init_trace_selftests);
2139 #else
run_tracer_selftest(struct tracer * type)2140 static inline int run_tracer_selftest(struct tracer *type)
2141 {
2142 return 0;
2143 }
do_run_tracer_selftest(struct tracer * type)2144 static inline int do_run_tracer_selftest(struct tracer *type)
2145 {
2146 return 0;
2147 }
2148 #endif /* CONFIG_FTRACE_STARTUP_TEST */
2149
2150 static void add_tracer_options(struct trace_array *tr, struct tracer *t);
2151
2152 static void __init apply_trace_boot_options(void);
2153
2154 /**
2155 * register_tracer - register a tracer with the ftrace system.
2156 * @type: the plugin for the tracer
2157 *
2158 * Register a new plugin tracer.
2159 */
register_tracer(struct tracer * type)2160 int __init register_tracer(struct tracer *type)
2161 {
2162 struct tracer *t;
2163 int ret = 0;
2164
2165 if (!type->name) {
2166 pr_info("Tracer must have a name\n");
2167 return -1;
2168 }
2169
2170 if (strlen(type->name) >= MAX_TRACER_SIZE) {
2171 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
2172 return -1;
2173 }
2174
2175 if (security_locked_down(LOCKDOWN_TRACEFS)) {
2176 pr_warn("Can not register tracer %s due to lockdown\n",
2177 type->name);
2178 return -EPERM;
2179 }
2180
2181 mutex_lock(&trace_types_lock);
2182
2183 for (t = trace_types; t; t = t->next) {
2184 if (strcmp(type->name, t->name) == 0) {
2185 /* already found */
2186 pr_info("Tracer %s already registered\n",
2187 type->name);
2188 ret = -1;
2189 goto out;
2190 }
2191 }
2192
2193 if (!type->set_flag)
2194 type->set_flag = &dummy_set_flag;
2195 if (!type->flags) {
2196 /*allocate a dummy tracer_flags*/
2197 type->flags = kmalloc(sizeof(*type->flags), GFP_KERNEL);
2198 if (!type->flags) {
2199 ret = -ENOMEM;
2200 goto out;
2201 }
2202 type->flags->val = 0;
2203 type->flags->opts = dummy_tracer_opt;
2204 } else
2205 if (!type->flags->opts)
2206 type->flags->opts = dummy_tracer_opt;
2207
2208 /* store the tracer for __set_tracer_option */
2209 type->flags->trace = type;
2210
2211 ret = do_run_tracer_selftest(type);
2212 if (ret < 0)
2213 goto out;
2214
2215 type->next = trace_types;
2216 trace_types = type;
2217 add_tracer_options(&global_trace, type);
2218
2219 out:
2220 mutex_unlock(&trace_types_lock);
2221
2222 if (ret || !default_bootup_tracer)
2223 goto out_unlock;
2224
2225 if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
2226 goto out_unlock;
2227
2228 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
2229 /* Do we want this tracer to start on bootup? */
2230 tracing_set_tracer(&global_trace, type->name);
2231 default_bootup_tracer = NULL;
2232
2233 apply_trace_boot_options();
2234
2235 /* disable other selftests, since this will break it. */
2236 disable_tracing_selftest("running a tracer");
2237
2238 out_unlock:
2239 return ret;
2240 }
2241
tracing_reset_cpu(struct array_buffer * buf,int cpu)2242 static void tracing_reset_cpu(struct array_buffer *buf, int cpu)
2243 {
2244 struct trace_buffer *buffer = buf->buffer;
2245
2246 if (!buffer)
2247 return;
2248
2249 ring_buffer_record_disable(buffer);
2250
2251 /* Make sure all commits have finished */
2252 synchronize_rcu();
2253 ring_buffer_reset_cpu(buffer, cpu);
2254
2255 ring_buffer_record_enable(buffer);
2256 }
2257
tracing_reset_online_cpus(struct array_buffer * buf)2258 void tracing_reset_online_cpus(struct array_buffer *buf)
2259 {
2260 struct trace_buffer *buffer = buf->buffer;
2261
2262 if (!buffer)
2263 return;
2264
2265 ring_buffer_record_disable(buffer);
2266
2267 /* Make sure all commits have finished */
2268 synchronize_rcu();
2269
2270 buf->time_start = buffer_ftrace_now(buf, buf->cpu);
2271
2272 ring_buffer_reset_online_cpus(buffer);
2273
2274 ring_buffer_record_enable(buffer);
2275 }
2276
2277 /* Must have trace_types_lock held */
tracing_reset_all_online_cpus_unlocked(void)2278 void tracing_reset_all_online_cpus_unlocked(void)
2279 {
2280 struct trace_array *tr;
2281
2282 lockdep_assert_held(&trace_types_lock);
2283
2284 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
2285 if (!tr->clear_trace)
2286 continue;
2287 tr->clear_trace = false;
2288 tracing_reset_online_cpus(&tr->array_buffer);
2289 #ifdef CONFIG_TRACER_MAX_TRACE
2290 tracing_reset_online_cpus(&tr->max_buffer);
2291 #endif
2292 }
2293 }
2294
tracing_reset_all_online_cpus(void)2295 void tracing_reset_all_online_cpus(void)
2296 {
2297 mutex_lock(&trace_types_lock);
2298 tracing_reset_all_online_cpus_unlocked();
2299 mutex_unlock(&trace_types_lock);
2300 }
2301
2302 /*
2303 * The tgid_map array maps from pid to tgid; i.e. the value stored at index i
2304 * is the tgid last observed corresponding to pid=i.
2305 */
2306 static int *tgid_map;
2307
2308 /* The maximum valid index into tgid_map. */
2309 static size_t tgid_map_max;
2310
2311 #define SAVED_CMDLINES_DEFAULT 128
2312 #define NO_CMDLINE_MAP UINT_MAX
2313 /*
2314 * Preemption must be disabled before acquiring trace_cmdline_lock.
2315 * The various trace_arrays' max_lock must be acquired in a context
2316 * where interrupt is disabled.
2317 */
2318 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
2319 struct saved_cmdlines_buffer {
2320 unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
2321 unsigned *map_cmdline_to_pid;
2322 unsigned cmdline_num;
2323 int cmdline_idx;
2324 char saved_cmdlines[];
2325 };
2326 static struct saved_cmdlines_buffer *savedcmd;
2327
get_saved_cmdlines(int idx)2328 static inline char *get_saved_cmdlines(int idx)
2329 {
2330 return &savedcmd->saved_cmdlines[idx * TASK_COMM_LEN];
2331 }
2332
set_cmdline(int idx,const char * cmdline)2333 static inline void set_cmdline(int idx, const char *cmdline)
2334 {
2335 strncpy(get_saved_cmdlines(idx), cmdline, TASK_COMM_LEN);
2336 }
2337
free_saved_cmdlines_buffer(struct saved_cmdlines_buffer * s)2338 static void free_saved_cmdlines_buffer(struct saved_cmdlines_buffer *s)
2339 {
2340 int order = get_order(sizeof(*s) + s->cmdline_num * TASK_COMM_LEN);
2341
2342 kfree(s->map_cmdline_to_pid);
2343 kmemleak_free(s);
2344 free_pages((unsigned long)s, order);
2345 }
2346
allocate_cmdlines_buffer(unsigned int val)2347 static struct saved_cmdlines_buffer *allocate_cmdlines_buffer(unsigned int val)
2348 {
2349 struct saved_cmdlines_buffer *s;
2350 struct page *page;
2351 int orig_size, size;
2352 int order;
2353
2354 /* Figure out how much is needed to hold the given number of cmdlines */
2355 orig_size = sizeof(*s) + val * TASK_COMM_LEN;
2356 order = get_order(orig_size);
2357 size = 1 << (order + PAGE_SHIFT);
2358 page = alloc_pages(GFP_KERNEL, order);
2359 if (!page)
2360 return NULL;
2361
2362 s = page_address(page);
2363 kmemleak_alloc(s, size, 1, GFP_KERNEL);
2364 memset(s, 0, sizeof(*s));
2365
2366 /* Round up to actual allocation */
2367 val = (size - sizeof(*s)) / TASK_COMM_LEN;
2368 s->cmdline_num = val;
2369
2370 s->map_cmdline_to_pid = kmalloc_array(val,
2371 sizeof(*s->map_cmdline_to_pid),
2372 GFP_KERNEL);
2373 if (!s->map_cmdline_to_pid) {
2374 free_saved_cmdlines_buffer(s);
2375 return NULL;
2376 }
2377
2378 s->cmdline_idx = 0;
2379 memset(&s->map_pid_to_cmdline, NO_CMDLINE_MAP,
2380 sizeof(s->map_pid_to_cmdline));
2381 memset(s->map_cmdline_to_pid, NO_CMDLINE_MAP,
2382 val * sizeof(*s->map_cmdline_to_pid));
2383
2384 return s;
2385 }
2386
trace_create_savedcmd(void)2387 static int trace_create_savedcmd(void)
2388 {
2389 savedcmd = allocate_cmdlines_buffer(SAVED_CMDLINES_DEFAULT);
2390
2391 return savedcmd ? 0 : -ENOMEM;
2392 }
2393
is_tracing_stopped(void)2394 int is_tracing_stopped(void)
2395 {
2396 return global_trace.stop_count;
2397 }
2398
tracing_start_tr(struct trace_array * tr)2399 static void tracing_start_tr(struct trace_array *tr)
2400 {
2401 struct trace_buffer *buffer;
2402 unsigned long flags;
2403
2404 if (tracing_disabled)
2405 return;
2406
2407 raw_spin_lock_irqsave(&tr->start_lock, flags);
2408 if (--tr->stop_count) {
2409 if (WARN_ON_ONCE(tr->stop_count < 0)) {
2410 /* Someone screwed up their debugging */
2411 tr->stop_count = 0;
2412 }
2413 goto out;
2414 }
2415
2416 /* Prevent the buffers from switching */
2417 arch_spin_lock(&tr->max_lock);
2418
2419 buffer = tr->array_buffer.buffer;
2420 if (buffer)
2421 ring_buffer_record_enable(buffer);
2422
2423 #ifdef CONFIG_TRACER_MAX_TRACE
2424 buffer = tr->max_buffer.buffer;
2425 if (buffer)
2426 ring_buffer_record_enable(buffer);
2427 #endif
2428
2429 arch_spin_unlock(&tr->max_lock);
2430
2431 out:
2432 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2433 }
2434
2435 /**
2436 * tracing_start - quick start of the tracer
2437 *
2438 * If tracing is enabled but was stopped by tracing_stop,
2439 * this will start the tracer back up.
2440 */
tracing_start(void)2441 void tracing_start(void)
2442
2443 {
2444 return tracing_start_tr(&global_trace);
2445 }
2446
tracing_stop_tr(struct trace_array * tr)2447 static void tracing_stop_tr(struct trace_array *tr)
2448 {
2449 struct trace_buffer *buffer;
2450 unsigned long flags;
2451
2452 raw_spin_lock_irqsave(&tr->start_lock, flags);
2453 if (tr->stop_count++)
2454 goto out;
2455
2456 /* Prevent the buffers from switching */
2457 arch_spin_lock(&tr->max_lock);
2458
2459 buffer = tr->array_buffer.buffer;
2460 if (buffer)
2461 ring_buffer_record_disable(buffer);
2462
2463 #ifdef CONFIG_TRACER_MAX_TRACE
2464 buffer = tr->max_buffer.buffer;
2465 if (buffer)
2466 ring_buffer_record_disable(buffer);
2467 #endif
2468
2469 arch_spin_unlock(&tr->max_lock);
2470
2471 out:
2472 raw_spin_unlock_irqrestore(&tr->start_lock, flags);
2473 }
2474
2475 /**
2476 * tracing_stop - quick stop of the tracer
2477 *
2478 * Light weight way to stop tracing. Use in conjunction with
2479 * tracing_start.
2480 */
tracing_stop(void)2481 void tracing_stop(void)
2482 {
2483 return tracing_stop_tr(&global_trace);
2484 }
2485
trace_save_cmdline(struct task_struct * tsk)2486 static int trace_save_cmdline(struct task_struct *tsk)
2487 {
2488 unsigned tpid, idx;
2489
2490 /* treat recording of idle task as a success */
2491 if (!tsk->pid)
2492 return 1;
2493
2494 tpid = tsk->pid & (PID_MAX_DEFAULT - 1);
2495
2496 /*
2497 * It's not the end of the world if we don't get
2498 * the lock, but we also don't want to spin
2499 * nor do we want to disable interrupts,
2500 * so if we miss here, then better luck next time.
2501 *
2502 * This is called within the scheduler and wake up, so interrupts
2503 * had better been disabled and run queue lock been held.
2504 */
2505 lockdep_assert_preemption_disabled();
2506 if (!arch_spin_trylock(&trace_cmdline_lock))
2507 return 0;
2508
2509 idx = savedcmd->map_pid_to_cmdline[tpid];
2510 if (idx == NO_CMDLINE_MAP) {
2511 idx = (savedcmd->cmdline_idx + 1) % savedcmd->cmdline_num;
2512
2513 savedcmd->map_pid_to_cmdline[tpid] = idx;
2514 savedcmd->cmdline_idx = idx;
2515 }
2516
2517 savedcmd->map_cmdline_to_pid[idx] = tsk->pid;
2518 set_cmdline(idx, tsk->comm);
2519
2520 arch_spin_unlock(&trace_cmdline_lock);
2521
2522 return 1;
2523 }
2524
__trace_find_cmdline(int pid,char comm[])2525 static void __trace_find_cmdline(int pid, char comm[])
2526 {
2527 unsigned map;
2528 int tpid;
2529
2530 if (!pid) {
2531 strcpy(comm, "<idle>");
2532 return;
2533 }
2534
2535 if (WARN_ON_ONCE(pid < 0)) {
2536 strcpy(comm, "<XXX>");
2537 return;
2538 }
2539
2540 tpid = pid & (PID_MAX_DEFAULT - 1);
2541 map = savedcmd->map_pid_to_cmdline[tpid];
2542 if (map != NO_CMDLINE_MAP) {
2543 tpid = savedcmd->map_cmdline_to_pid[map];
2544 if (tpid == pid) {
2545 strscpy(comm, get_saved_cmdlines(map), TASK_COMM_LEN);
2546 return;
2547 }
2548 }
2549 strcpy(comm, "<...>");
2550 }
2551
trace_find_cmdline(int pid,char comm[])2552 void trace_find_cmdline(int pid, char comm[])
2553 {
2554 preempt_disable();
2555 arch_spin_lock(&trace_cmdline_lock);
2556
2557 __trace_find_cmdline(pid, comm);
2558
2559 arch_spin_unlock(&trace_cmdline_lock);
2560 preempt_enable();
2561 }
2562
trace_find_tgid_ptr(int pid)2563 static int *trace_find_tgid_ptr(int pid)
2564 {
2565 /*
2566 * Pairs with the smp_store_release in set_tracer_flag() to ensure that
2567 * if we observe a non-NULL tgid_map then we also observe the correct
2568 * tgid_map_max.
2569 */
2570 int *map = smp_load_acquire(&tgid_map);
2571
2572 if (unlikely(!map || pid > tgid_map_max))
2573 return NULL;
2574
2575 return &map[pid];
2576 }
2577
trace_find_tgid(int pid)2578 int trace_find_tgid(int pid)
2579 {
2580 int *ptr = trace_find_tgid_ptr(pid);
2581
2582 return ptr ? *ptr : 0;
2583 }
2584
trace_save_tgid(struct task_struct * tsk)2585 static int trace_save_tgid(struct task_struct *tsk)
2586 {
2587 int *ptr;
2588
2589 /* treat recording of idle task as a success */
2590 if (!tsk->pid)
2591 return 1;
2592
2593 ptr = trace_find_tgid_ptr(tsk->pid);
2594 if (!ptr)
2595 return 0;
2596
2597 *ptr = tsk->tgid;
2598 return 1;
2599 }
2600
tracing_record_taskinfo_skip(int flags)2601 static bool tracing_record_taskinfo_skip(int flags)
2602 {
2603 if (unlikely(!(flags & (TRACE_RECORD_CMDLINE | TRACE_RECORD_TGID))))
2604 return true;
2605 if (!__this_cpu_read(trace_taskinfo_save))
2606 return true;
2607 return false;
2608 }
2609
2610 /**
2611 * tracing_record_taskinfo - record the task info of a task
2612 *
2613 * @task: task to record
2614 * @flags: TRACE_RECORD_CMDLINE for recording comm
2615 * TRACE_RECORD_TGID for recording tgid
2616 */
tracing_record_taskinfo(struct task_struct * task,int flags)2617 void tracing_record_taskinfo(struct task_struct *task, int flags)
2618 {
2619 bool done;
2620
2621 if (tracing_record_taskinfo_skip(flags))
2622 return;
2623
2624 /*
2625 * Record as much task information as possible. If some fail, continue
2626 * to try to record the others.
2627 */
2628 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(task);
2629 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(task);
2630
2631 /* If recording any information failed, retry again soon. */
2632 if (!done)
2633 return;
2634
2635 __this_cpu_write(trace_taskinfo_save, false);
2636 }
2637
2638 /**
2639 * tracing_record_taskinfo_sched_switch - record task info for sched_switch
2640 *
2641 * @prev: previous task during sched_switch
2642 * @next: next task during sched_switch
2643 * @flags: TRACE_RECORD_CMDLINE for recording comm
2644 * TRACE_RECORD_TGID for recording tgid
2645 */
tracing_record_taskinfo_sched_switch(struct task_struct * prev,struct task_struct * next,int flags)2646 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
2647 struct task_struct *next, int flags)
2648 {
2649 bool done;
2650
2651 if (tracing_record_taskinfo_skip(flags))
2652 return;
2653
2654 /*
2655 * Record as much task information as possible. If some fail, continue
2656 * to try to record the others.
2657 */
2658 done = !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(prev);
2659 done &= !(flags & TRACE_RECORD_CMDLINE) || trace_save_cmdline(next);
2660 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(prev);
2661 done &= !(flags & TRACE_RECORD_TGID) || trace_save_tgid(next);
2662
2663 /* If recording any information failed, retry again soon. */
2664 if (!done)
2665 return;
2666
2667 __this_cpu_write(trace_taskinfo_save, false);
2668 }
2669
2670 /* Helpers to record a specific task information */
tracing_record_cmdline(struct task_struct * task)2671 void tracing_record_cmdline(struct task_struct *task)
2672 {
2673 tracing_record_taskinfo(task, TRACE_RECORD_CMDLINE);
2674 }
2675
tracing_record_tgid(struct task_struct * task)2676 void tracing_record_tgid(struct task_struct *task)
2677 {
2678 tracing_record_taskinfo(task, TRACE_RECORD_TGID);
2679 }
2680
2681 /*
2682 * Several functions return TRACE_TYPE_PARTIAL_LINE if the trace_seq
2683 * overflowed, and TRACE_TYPE_HANDLED otherwise. This helper function
2684 * simplifies those functions and keeps them in sync.
2685 */
trace_handle_return(struct trace_seq * s)2686 enum print_line_t trace_handle_return(struct trace_seq *s)
2687 {
2688 return trace_seq_has_overflowed(s) ?
2689 TRACE_TYPE_PARTIAL_LINE : TRACE_TYPE_HANDLED;
2690 }
2691 EXPORT_SYMBOL_GPL(trace_handle_return);
2692
migration_disable_value(void)2693 static unsigned short migration_disable_value(void)
2694 {
2695 #if defined(CONFIG_SMP)
2696 return current->migration_disabled;
2697 #else
2698 return 0;
2699 #endif
2700 }
2701
tracing_gen_ctx_irq_test(unsigned int irqs_status)2702 unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status)
2703 {
2704 unsigned int trace_flags = irqs_status;
2705 unsigned int pc;
2706
2707 pc = preempt_count();
2708
2709 if (pc & NMI_MASK)
2710 trace_flags |= TRACE_FLAG_NMI;
2711 if (pc & HARDIRQ_MASK)
2712 trace_flags |= TRACE_FLAG_HARDIRQ;
2713 if (in_serving_softirq())
2714 trace_flags |= TRACE_FLAG_SOFTIRQ;
2715 if (softirq_count() >> (SOFTIRQ_SHIFT + 1))
2716 trace_flags |= TRACE_FLAG_BH_OFF;
2717
2718 if (tif_need_resched())
2719 trace_flags |= TRACE_FLAG_NEED_RESCHED;
2720 if (test_preempt_need_resched())
2721 trace_flags |= TRACE_FLAG_PREEMPT_RESCHED;
2722 return (trace_flags << 16) | (min_t(unsigned int, pc & 0xff, 0xf)) |
2723 (min_t(unsigned int, migration_disable_value(), 0xf)) << 4;
2724 }
2725
2726 struct ring_buffer_event *
trace_buffer_lock_reserve(struct trace_buffer * buffer,int type,unsigned long len,unsigned int trace_ctx)2727 trace_buffer_lock_reserve(struct trace_buffer *buffer,
2728 int type,
2729 unsigned long len,
2730 unsigned int trace_ctx)
2731 {
2732 return __trace_buffer_lock_reserve(buffer, type, len, trace_ctx);
2733 }
2734
2735 DEFINE_PER_CPU(struct ring_buffer_event *, trace_buffered_event);
2736 DEFINE_PER_CPU(int, trace_buffered_event_cnt);
2737 static int trace_buffered_event_ref;
2738
2739 /**
2740 * trace_buffered_event_enable - enable buffering events
2741 *
2742 * When events are being filtered, it is quicker to use a temporary
2743 * buffer to write the event data into if there's a likely chance
2744 * that it will not be committed. The discard of the ring buffer
2745 * is not as fast as committing, and is much slower than copying
2746 * a commit.
2747 *
2748 * When an event is to be filtered, allocate per cpu buffers to
2749 * write the event data into, and if the event is filtered and discarded
2750 * it is simply dropped, otherwise, the entire data is to be committed
2751 * in one shot.
2752 */
trace_buffered_event_enable(void)2753 void trace_buffered_event_enable(void)
2754 {
2755 struct ring_buffer_event *event;
2756 struct page *page;
2757 int cpu;
2758
2759 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2760
2761 if (trace_buffered_event_ref++)
2762 return;
2763
2764 for_each_tracing_cpu(cpu) {
2765 page = alloc_pages_node(cpu_to_node(cpu),
2766 GFP_KERNEL | __GFP_NORETRY, 0);
2767 /* This is just an optimization and can handle failures */
2768 if (!page) {
2769 pr_err("Failed to allocate event buffer\n");
2770 break;
2771 }
2772
2773 event = page_address(page);
2774 memset(event, 0, sizeof(*event));
2775
2776 per_cpu(trace_buffered_event, cpu) = event;
2777
2778 preempt_disable();
2779 if (cpu == smp_processor_id() &&
2780 __this_cpu_read(trace_buffered_event) !=
2781 per_cpu(trace_buffered_event, cpu))
2782 WARN_ON_ONCE(1);
2783 preempt_enable();
2784 }
2785 }
2786
enable_trace_buffered_event(void * data)2787 static void enable_trace_buffered_event(void *data)
2788 {
2789 /* Probably not needed, but do it anyway */
2790 smp_rmb();
2791 this_cpu_dec(trace_buffered_event_cnt);
2792 }
2793
disable_trace_buffered_event(void * data)2794 static void disable_trace_buffered_event(void *data)
2795 {
2796 this_cpu_inc(trace_buffered_event_cnt);
2797 }
2798
2799 /**
2800 * trace_buffered_event_disable - disable buffering events
2801 *
2802 * When a filter is removed, it is faster to not use the buffered
2803 * events, and to commit directly into the ring buffer. Free up
2804 * the temp buffers when there are no more users. This requires
2805 * special synchronization with current events.
2806 */
trace_buffered_event_disable(void)2807 void trace_buffered_event_disable(void)
2808 {
2809 int cpu;
2810
2811 WARN_ON_ONCE(!mutex_is_locked(&event_mutex));
2812
2813 if (WARN_ON_ONCE(!trace_buffered_event_ref))
2814 return;
2815
2816 if (--trace_buffered_event_ref)
2817 return;
2818
2819 /* For each CPU, set the buffer as used. */
2820 on_each_cpu_mask(tracing_buffer_mask, disable_trace_buffered_event,
2821 NULL, true);
2822
2823 /* Wait for all current users to finish */
2824 synchronize_rcu();
2825
2826 for_each_tracing_cpu(cpu) {
2827 free_page((unsigned long)per_cpu(trace_buffered_event, cpu));
2828 per_cpu(trace_buffered_event, cpu) = NULL;
2829 }
2830
2831 /*
2832 * Wait for all CPUs that potentially started checking if they can use
2833 * their event buffer only after the previous synchronize_rcu() call and
2834 * they still read a valid pointer from trace_buffered_event. It must be
2835 * ensured they don't see cleared trace_buffered_event_cnt else they
2836 * could wrongly decide to use the pointed-to buffer which is now freed.
2837 */
2838 synchronize_rcu();
2839
2840 /* For each CPU, relinquish the buffer */
2841 on_each_cpu_mask(tracing_buffer_mask, enable_trace_buffered_event, NULL,
2842 true);
2843 }
2844
2845 static struct trace_buffer *temp_buffer;
2846
2847 struct ring_buffer_event *
trace_event_buffer_lock_reserve(struct trace_buffer ** current_rb,struct trace_event_file * trace_file,int type,unsigned long len,unsigned int trace_ctx)2848 trace_event_buffer_lock_reserve(struct trace_buffer **current_rb,
2849 struct trace_event_file *trace_file,
2850 int type, unsigned long len,
2851 unsigned int trace_ctx)
2852 {
2853 struct ring_buffer_event *entry;
2854 struct trace_array *tr = trace_file->tr;
2855 int val;
2856
2857 *current_rb = tr->array_buffer.buffer;
2858
2859 if (!tr->no_filter_buffering_ref &&
2860 (trace_file->flags & (EVENT_FILE_FL_SOFT_DISABLED | EVENT_FILE_FL_FILTERED))) {
2861 preempt_disable_notrace();
2862 /*
2863 * Filtering is on, so try to use the per cpu buffer first.
2864 * This buffer will simulate a ring_buffer_event,
2865 * where the type_len is zero and the array[0] will
2866 * hold the full length.
2867 * (see include/linux/ring-buffer.h for details on
2868 * how the ring_buffer_event is structured).
2869 *
2870 * Using a temp buffer during filtering and copying it
2871 * on a matched filter is quicker than writing directly
2872 * into the ring buffer and then discarding it when
2873 * it doesn't match. That is because the discard
2874 * requires several atomic operations to get right.
2875 * Copying on match and doing nothing on a failed match
2876 * is still quicker than no copy on match, but having
2877 * to discard out of the ring buffer on a failed match.
2878 */
2879 if ((entry = __this_cpu_read(trace_buffered_event))) {
2880 int max_len = PAGE_SIZE - struct_size(entry, array, 1);
2881
2882 val = this_cpu_inc_return(trace_buffered_event_cnt);
2883
2884 /*
2885 * Preemption is disabled, but interrupts and NMIs
2886 * can still come in now. If that happens after
2887 * the above increment, then it will have to go
2888 * back to the old method of allocating the event
2889 * on the ring buffer, and if the filter fails, it
2890 * will have to call ring_buffer_discard_commit()
2891 * to remove it.
2892 *
2893 * Need to also check the unlikely case that the
2894 * length is bigger than the temp buffer size.
2895 * If that happens, then the reserve is pretty much
2896 * guaranteed to fail, as the ring buffer currently
2897 * only allows events less than a page. But that may
2898 * change in the future, so let the ring buffer reserve
2899 * handle the failure in that case.
2900 */
2901 if (val == 1 && likely(len <= max_len)) {
2902 trace_event_setup(entry, type, trace_ctx);
2903 entry->array[0] = len;
2904 /* Return with preemption disabled */
2905 return entry;
2906 }
2907 this_cpu_dec(trace_buffered_event_cnt);
2908 }
2909 /* __trace_buffer_lock_reserve() disables preemption */
2910 preempt_enable_notrace();
2911 }
2912
2913 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2914 trace_ctx);
2915 /*
2916 * If tracing is off, but we have triggers enabled
2917 * we still need to look at the event data. Use the temp_buffer
2918 * to store the trace event for the trigger to use. It's recursive
2919 * safe and will not be recorded anywhere.
2920 */
2921 if (!entry && trace_file->flags & EVENT_FILE_FL_TRIGGER_COND) {
2922 *current_rb = temp_buffer;
2923 entry = __trace_buffer_lock_reserve(*current_rb, type, len,
2924 trace_ctx);
2925 }
2926 return entry;
2927 }
2928 EXPORT_SYMBOL_GPL(trace_event_buffer_lock_reserve);
2929
2930 static DEFINE_RAW_SPINLOCK(tracepoint_iter_lock);
2931 static DEFINE_MUTEX(tracepoint_printk_mutex);
2932
output_printk(struct trace_event_buffer * fbuffer)2933 static void output_printk(struct trace_event_buffer *fbuffer)
2934 {
2935 struct trace_event_call *event_call;
2936 struct trace_event_file *file;
2937 struct trace_event *event;
2938 unsigned long flags;
2939 struct trace_iterator *iter = tracepoint_print_iter;
2940
2941 /* We should never get here if iter is NULL */
2942 if (WARN_ON_ONCE(!iter))
2943 return;
2944
2945 event_call = fbuffer->trace_file->event_call;
2946 if (!event_call || !event_call->event.funcs ||
2947 !event_call->event.funcs->trace)
2948 return;
2949
2950 file = fbuffer->trace_file;
2951 if (test_bit(EVENT_FILE_FL_SOFT_DISABLED_BIT, &file->flags) ||
2952 (unlikely(file->flags & EVENT_FILE_FL_FILTERED) &&
2953 !filter_match_preds(file->filter, fbuffer->entry)))
2954 return;
2955
2956 event = &fbuffer->trace_file->event_call->event;
2957
2958 raw_spin_lock_irqsave(&tracepoint_iter_lock, flags);
2959 trace_seq_init(&iter->seq);
2960 iter->ent = fbuffer->entry;
2961 event_call->event.funcs->trace(iter, 0, event);
2962 trace_seq_putc(&iter->seq, 0);
2963 printk("%s", iter->seq.buffer);
2964
2965 raw_spin_unlock_irqrestore(&tracepoint_iter_lock, flags);
2966 }
2967
tracepoint_printk_sysctl(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)2968 int tracepoint_printk_sysctl(struct ctl_table *table, int write,
2969 void *buffer, size_t *lenp,
2970 loff_t *ppos)
2971 {
2972 int save_tracepoint_printk;
2973 int ret;
2974
2975 mutex_lock(&tracepoint_printk_mutex);
2976 save_tracepoint_printk = tracepoint_printk;
2977
2978 ret = proc_dointvec(table, write, buffer, lenp, ppos);
2979
2980 /*
2981 * This will force exiting early, as tracepoint_printk
2982 * is always zero when tracepoint_printk_iter is not allocated
2983 */
2984 if (!tracepoint_print_iter)
2985 tracepoint_printk = 0;
2986
2987 if (save_tracepoint_printk == tracepoint_printk)
2988 goto out;
2989
2990 if (tracepoint_printk)
2991 static_key_enable(&tracepoint_printk_key.key);
2992 else
2993 static_key_disable(&tracepoint_printk_key.key);
2994
2995 out:
2996 mutex_unlock(&tracepoint_printk_mutex);
2997
2998 return ret;
2999 }
3000
trace_event_buffer_commit(struct trace_event_buffer * fbuffer)3001 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer)
3002 {
3003 enum event_trigger_type tt = ETT_NONE;
3004 struct trace_event_file *file = fbuffer->trace_file;
3005
3006 if (__event_trigger_test_discard(file, fbuffer->buffer, fbuffer->event,
3007 fbuffer->entry, &tt))
3008 goto discard;
3009
3010 if (static_key_false(&tracepoint_printk_key.key))
3011 output_printk(fbuffer);
3012
3013 if (static_branch_unlikely(&trace_event_exports_enabled))
3014 ftrace_exports(fbuffer->event, TRACE_EXPORT_EVENT);
3015
3016 trace_buffer_unlock_commit_regs(file->tr, fbuffer->buffer,
3017 fbuffer->event, fbuffer->trace_ctx, fbuffer->regs);
3018
3019 discard:
3020 if (tt)
3021 event_triggers_post_call(file, tt);
3022
3023 }
3024 EXPORT_SYMBOL_GPL(trace_event_buffer_commit);
3025
3026 /*
3027 * Skip 3:
3028 *
3029 * trace_buffer_unlock_commit_regs()
3030 * trace_event_buffer_commit()
3031 * trace_event_raw_event_xxx()
3032 */
3033 # define STACK_SKIP 3
3034
trace_buffer_unlock_commit_regs(struct trace_array * tr,struct trace_buffer * buffer,struct ring_buffer_event * event,unsigned int trace_ctx,struct pt_regs * regs)3035 void trace_buffer_unlock_commit_regs(struct trace_array *tr,
3036 struct trace_buffer *buffer,
3037 struct ring_buffer_event *event,
3038 unsigned int trace_ctx,
3039 struct pt_regs *regs)
3040 {
3041 __buffer_unlock_commit(buffer, event);
3042
3043 /*
3044 * If regs is not set, then skip the necessary functions.
3045 * Note, we can still get here via blktrace, wakeup tracer
3046 * and mmiotrace, but that's ok if they lose a function or
3047 * two. They are not that meaningful.
3048 */
3049 ftrace_trace_stack(tr, buffer, trace_ctx, regs ? 0 : STACK_SKIP, regs);
3050 ftrace_trace_userstack(tr, buffer, trace_ctx);
3051 }
3052
3053 /*
3054 * Similar to trace_buffer_unlock_commit_regs() but do not dump stack.
3055 */
3056 void
trace_buffer_unlock_commit_nostack(struct trace_buffer * buffer,struct ring_buffer_event * event)3057 trace_buffer_unlock_commit_nostack(struct trace_buffer *buffer,
3058 struct ring_buffer_event *event)
3059 {
3060 __buffer_unlock_commit(buffer, event);
3061 }
3062
3063 void
trace_function(struct trace_array * tr,unsigned long ip,unsigned long parent_ip,unsigned int trace_ctx)3064 trace_function(struct trace_array *tr, unsigned long ip, unsigned long
3065 parent_ip, unsigned int trace_ctx)
3066 {
3067 struct trace_event_call *call = &event_function;
3068 struct trace_buffer *buffer = tr->array_buffer.buffer;
3069 struct ring_buffer_event *event;
3070 struct ftrace_entry *entry;
3071
3072 event = __trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
3073 trace_ctx);
3074 if (!event)
3075 return;
3076 entry = ring_buffer_event_data(event);
3077 entry->ip = ip;
3078 entry->parent_ip = parent_ip;
3079
3080 if (!call_filter_check_discard(call, entry, buffer, event)) {
3081 if (static_branch_unlikely(&trace_function_exports_enabled))
3082 ftrace_exports(event, TRACE_EXPORT_FUNCTION);
3083 __buffer_unlock_commit(buffer, event);
3084 }
3085 }
3086
3087 #ifdef CONFIG_STACKTRACE
3088
3089 /* Allow 4 levels of nesting: normal, softirq, irq, NMI */
3090 #define FTRACE_KSTACK_NESTING 4
3091
3092 #define FTRACE_KSTACK_ENTRIES (PAGE_SIZE / FTRACE_KSTACK_NESTING)
3093
3094 struct ftrace_stack {
3095 unsigned long calls[FTRACE_KSTACK_ENTRIES];
3096 };
3097
3098
3099 struct ftrace_stacks {
3100 struct ftrace_stack stacks[FTRACE_KSTACK_NESTING];
3101 };
3102
3103 static DEFINE_PER_CPU(struct ftrace_stacks, ftrace_stacks);
3104 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
3105
__ftrace_trace_stack(struct trace_buffer * buffer,unsigned int trace_ctx,int skip,struct pt_regs * regs)3106 static void __ftrace_trace_stack(struct trace_buffer *buffer,
3107 unsigned int trace_ctx,
3108 int skip, struct pt_regs *regs)
3109 {
3110 struct trace_event_call *call = &event_kernel_stack;
3111 struct ring_buffer_event *event;
3112 unsigned int size, nr_entries;
3113 struct ftrace_stack *fstack;
3114 struct stack_entry *entry;
3115 int stackidx;
3116
3117 /*
3118 * Add one, for this function and the call to save_stack_trace()
3119 * If regs is set, then these functions will not be in the way.
3120 */
3121 #ifndef CONFIG_UNWINDER_ORC
3122 if (!regs)
3123 skip++;
3124 #endif
3125
3126 preempt_disable_notrace();
3127
3128 stackidx = __this_cpu_inc_return(ftrace_stack_reserve) - 1;
3129
3130 /* This should never happen. If it does, yell once and skip */
3131 if (WARN_ON_ONCE(stackidx >= FTRACE_KSTACK_NESTING))
3132 goto out;
3133
3134 /*
3135 * The above __this_cpu_inc_return() is 'atomic' cpu local. An
3136 * interrupt will either see the value pre increment or post
3137 * increment. If the interrupt happens pre increment it will have
3138 * restored the counter when it returns. We just need a barrier to
3139 * keep gcc from moving things around.
3140 */
3141 barrier();
3142
3143 fstack = this_cpu_ptr(ftrace_stacks.stacks) + stackidx;
3144 size = ARRAY_SIZE(fstack->calls);
3145
3146 if (regs) {
3147 nr_entries = stack_trace_save_regs(regs, fstack->calls,
3148 size, skip);
3149 } else {
3150 nr_entries = stack_trace_save(fstack->calls, size, skip);
3151 }
3152
3153 event = __trace_buffer_lock_reserve(buffer, TRACE_STACK,
3154 struct_size(entry, caller, nr_entries),
3155 trace_ctx);
3156 if (!event)
3157 goto out;
3158 entry = ring_buffer_event_data(event);
3159
3160 entry->size = nr_entries;
3161 memcpy(&entry->caller, fstack->calls,
3162 flex_array_size(entry, caller, nr_entries));
3163
3164 if (!call_filter_check_discard(call, entry, buffer, event))
3165 __buffer_unlock_commit(buffer, event);
3166
3167 out:
3168 /* Again, don't let gcc optimize things here */
3169 barrier();
3170 __this_cpu_dec(ftrace_stack_reserve);
3171 preempt_enable_notrace();
3172
3173 }
3174
ftrace_trace_stack(struct trace_array * tr,struct trace_buffer * buffer,unsigned int trace_ctx,int skip,struct pt_regs * regs)3175 static inline void ftrace_trace_stack(struct trace_array *tr,
3176 struct trace_buffer *buffer,
3177 unsigned int trace_ctx,
3178 int skip, struct pt_regs *regs)
3179 {
3180 if (!(tr->trace_flags & TRACE_ITER_STACKTRACE))
3181 return;
3182
3183 __ftrace_trace_stack(buffer, trace_ctx, skip, regs);
3184 }
3185
__trace_stack(struct trace_array * tr,unsigned int trace_ctx,int skip)3186 void __trace_stack(struct trace_array *tr, unsigned int trace_ctx,
3187 int skip)
3188 {
3189 struct trace_buffer *buffer = tr->array_buffer.buffer;
3190
3191 if (rcu_is_watching()) {
3192 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3193 return;
3194 }
3195
3196 if (WARN_ON_ONCE(IS_ENABLED(CONFIG_GENERIC_ENTRY)))
3197 return;
3198
3199 /*
3200 * When an NMI triggers, RCU is enabled via ct_nmi_enter(),
3201 * but if the above rcu_is_watching() failed, then the NMI
3202 * triggered someplace critical, and ct_irq_enter() should
3203 * not be called from NMI.
3204 */
3205 if (unlikely(in_nmi()))
3206 return;
3207
3208 ct_irq_enter_irqson();
3209 __ftrace_trace_stack(buffer, trace_ctx, skip, NULL);
3210 ct_irq_exit_irqson();
3211 }
3212
3213 /**
3214 * trace_dump_stack - record a stack back trace in the trace buffer
3215 * @skip: Number of functions to skip (helper handlers)
3216 */
trace_dump_stack(int skip)3217 void trace_dump_stack(int skip)
3218 {
3219 if (tracing_disabled || tracing_selftest_running)
3220 return;
3221
3222 #ifndef CONFIG_UNWINDER_ORC
3223 /* Skip 1 to skip this function. */
3224 skip++;
3225 #endif
3226 __ftrace_trace_stack(global_trace.array_buffer.buffer,
3227 tracing_gen_ctx(), skip, NULL);
3228 }
3229 EXPORT_SYMBOL_GPL(trace_dump_stack);
3230
3231 #ifdef CONFIG_USER_STACKTRACE_SUPPORT
3232 static DEFINE_PER_CPU(int, user_stack_count);
3233
3234 static void
ftrace_trace_userstack(struct trace_array * tr,struct trace_buffer * buffer,unsigned int trace_ctx)3235 ftrace_trace_userstack(struct trace_array *tr,
3236 struct trace_buffer *buffer, unsigned int trace_ctx)
3237 {
3238 struct trace_event_call *call = &event_user_stack;
3239 struct ring_buffer_event *event;
3240 struct userstack_entry *entry;
3241
3242 if (!(tr->trace_flags & TRACE_ITER_USERSTACKTRACE))
3243 return;
3244
3245 /*
3246 * NMIs can not handle page faults, even with fix ups.
3247 * The save user stack can (and often does) fault.
3248 */
3249 if (unlikely(in_nmi()))
3250 return;
3251
3252 /*
3253 * prevent recursion, since the user stack tracing may
3254 * trigger other kernel events.
3255 */
3256 preempt_disable();
3257 if (__this_cpu_read(user_stack_count))
3258 goto out;
3259
3260 __this_cpu_inc(user_stack_count);
3261
3262 event = __trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
3263 sizeof(*entry), trace_ctx);
3264 if (!event)
3265 goto out_drop_count;
3266 entry = ring_buffer_event_data(event);
3267
3268 entry->tgid = current->tgid;
3269 memset(&entry->caller, 0, sizeof(entry->caller));
3270
3271 stack_trace_save_user(entry->caller, FTRACE_STACK_ENTRIES);
3272 if (!call_filter_check_discard(call, entry, buffer, event))
3273 __buffer_unlock_commit(buffer, event);
3274
3275 out_drop_count:
3276 __this_cpu_dec(user_stack_count);
3277 out:
3278 preempt_enable();
3279 }
3280 #else /* CONFIG_USER_STACKTRACE_SUPPORT */
ftrace_trace_userstack(struct trace_array * tr,struct trace_buffer * buffer,unsigned int trace_ctx)3281 static void ftrace_trace_userstack(struct trace_array *tr,
3282 struct trace_buffer *buffer,
3283 unsigned int trace_ctx)
3284 {
3285 }
3286 #endif /* !CONFIG_USER_STACKTRACE_SUPPORT */
3287
3288 #endif /* CONFIG_STACKTRACE */
3289
3290 static inline void
func_repeats_set_delta_ts(struct func_repeats_entry * entry,unsigned long long delta)3291 func_repeats_set_delta_ts(struct func_repeats_entry *entry,
3292 unsigned long long delta)
3293 {
3294 entry->bottom_delta_ts = delta & U32_MAX;
3295 entry->top_delta_ts = (delta >> 32);
3296 }
3297
trace_last_func_repeats(struct trace_array * tr,struct trace_func_repeats * last_info,unsigned int trace_ctx)3298 void trace_last_func_repeats(struct trace_array *tr,
3299 struct trace_func_repeats *last_info,
3300 unsigned int trace_ctx)
3301 {
3302 struct trace_buffer *buffer = tr->array_buffer.buffer;
3303 struct func_repeats_entry *entry;
3304 struct ring_buffer_event *event;
3305 u64 delta;
3306
3307 event = __trace_buffer_lock_reserve(buffer, TRACE_FUNC_REPEATS,
3308 sizeof(*entry), trace_ctx);
3309 if (!event)
3310 return;
3311
3312 delta = ring_buffer_event_time_stamp(buffer, event) -
3313 last_info->ts_last_call;
3314
3315 entry = ring_buffer_event_data(event);
3316 entry->ip = last_info->ip;
3317 entry->parent_ip = last_info->parent_ip;
3318 entry->count = last_info->count;
3319 func_repeats_set_delta_ts(entry, delta);
3320
3321 __buffer_unlock_commit(buffer, event);
3322 }
3323
3324 /* created for use with alloc_percpu */
3325 struct trace_buffer_struct {
3326 int nesting;
3327 char buffer[4][TRACE_BUF_SIZE];
3328 };
3329
3330 static struct trace_buffer_struct __percpu *trace_percpu_buffer;
3331
3332 /*
3333 * This allows for lockless recording. If we're nested too deeply, then
3334 * this returns NULL.
3335 */
get_trace_buf(void)3336 static char *get_trace_buf(void)
3337 {
3338 struct trace_buffer_struct *buffer = this_cpu_ptr(trace_percpu_buffer);
3339
3340 if (!trace_percpu_buffer || buffer->nesting >= 4)
3341 return NULL;
3342
3343 buffer->nesting++;
3344
3345 /* Interrupts must see nesting incremented before we use the buffer */
3346 barrier();
3347 return &buffer->buffer[buffer->nesting - 1][0];
3348 }
3349
put_trace_buf(void)3350 static void put_trace_buf(void)
3351 {
3352 /* Don't let the decrement of nesting leak before this */
3353 barrier();
3354 this_cpu_dec(trace_percpu_buffer->nesting);
3355 }
3356
alloc_percpu_trace_buffer(void)3357 static int alloc_percpu_trace_buffer(void)
3358 {
3359 struct trace_buffer_struct __percpu *buffers;
3360
3361 if (trace_percpu_buffer)
3362 return 0;
3363
3364 buffers = alloc_percpu(struct trace_buffer_struct);
3365 if (MEM_FAIL(!buffers, "Could not allocate percpu trace_printk buffer"))
3366 return -ENOMEM;
3367
3368 trace_percpu_buffer = buffers;
3369 return 0;
3370 }
3371
3372 static int buffers_allocated;
3373
trace_printk_init_buffers(void)3374 void trace_printk_init_buffers(void)
3375 {
3376 if (buffers_allocated)
3377 return;
3378
3379 if (alloc_percpu_trace_buffer())
3380 return;
3381
3382 /* trace_printk() is for debug use only. Don't use it in production. */
3383
3384 pr_warn("\n");
3385 pr_warn("**********************************************************\n");
3386 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3387 pr_warn("** **\n");
3388 pr_warn("** trace_printk() being used. Allocating extra memory. **\n");
3389 pr_warn("** **\n");
3390 pr_warn("** This means that this is a DEBUG kernel and it is **\n");
3391 pr_warn("** unsafe for production use. **\n");
3392 pr_warn("** **\n");
3393 pr_warn("** If you see this message and you are not debugging **\n");
3394 pr_warn("** the kernel, report this immediately to your vendor! **\n");
3395 pr_warn("** **\n");
3396 pr_warn("** NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE NOTICE **\n");
3397 pr_warn("**********************************************************\n");
3398
3399 /* Expand the buffers to set size */
3400 tracing_update_buffers(&global_trace);
3401
3402 buffers_allocated = 1;
3403
3404 /*
3405 * trace_printk_init_buffers() can be called by modules.
3406 * If that happens, then we need to start cmdline recording
3407 * directly here. If the global_trace.buffer is already
3408 * allocated here, then this was called by module code.
3409 */
3410 if (global_trace.array_buffer.buffer)
3411 tracing_start_cmdline_record();
3412 }
3413 EXPORT_SYMBOL_GPL(trace_printk_init_buffers);
3414
trace_printk_start_comm(void)3415 void trace_printk_start_comm(void)
3416 {
3417 /* Start tracing comms if trace printk is set */
3418 if (!buffers_allocated)
3419 return;
3420 tracing_start_cmdline_record();
3421 }
3422
trace_printk_start_stop_comm(int enabled)3423 static void trace_printk_start_stop_comm(int enabled)
3424 {
3425 if (!buffers_allocated)
3426 return;
3427
3428 if (enabled)
3429 tracing_start_cmdline_record();
3430 else
3431 tracing_stop_cmdline_record();
3432 }
3433
3434 /**
3435 * trace_vbprintk - write binary msg to tracing buffer
3436 * @ip: The address of the caller
3437 * @fmt: The string format to write to the buffer
3438 * @args: Arguments for @fmt
3439 */
trace_vbprintk(unsigned long ip,const char * fmt,va_list args)3440 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
3441 {
3442 struct trace_event_call *call = &event_bprint;
3443 struct ring_buffer_event *event;
3444 struct trace_buffer *buffer;
3445 struct trace_array *tr = &global_trace;
3446 struct bprint_entry *entry;
3447 unsigned int trace_ctx;
3448 char *tbuffer;
3449 int len = 0, size;
3450
3451 if (unlikely(tracing_selftest_running || tracing_disabled))
3452 return 0;
3453
3454 /* Don't pollute graph traces with trace_vprintk internals */
3455 pause_graph_tracing();
3456
3457 trace_ctx = tracing_gen_ctx();
3458 preempt_disable_notrace();
3459
3460 tbuffer = get_trace_buf();
3461 if (!tbuffer) {
3462 len = 0;
3463 goto out_nobuffer;
3464 }
3465
3466 len = vbin_printf((u32 *)tbuffer, TRACE_BUF_SIZE/sizeof(int), fmt, args);
3467
3468 if (len > TRACE_BUF_SIZE/sizeof(int) || len < 0)
3469 goto out_put;
3470
3471 size = sizeof(*entry) + sizeof(u32) * len;
3472 buffer = tr->array_buffer.buffer;
3473 ring_buffer_nest_start(buffer);
3474 event = __trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
3475 trace_ctx);
3476 if (!event)
3477 goto out;
3478 entry = ring_buffer_event_data(event);
3479 entry->ip = ip;
3480 entry->fmt = fmt;
3481
3482 memcpy(entry->buf, tbuffer, sizeof(u32) * len);
3483 if (!call_filter_check_discard(call, entry, buffer, event)) {
3484 __buffer_unlock_commit(buffer, event);
3485 ftrace_trace_stack(tr, buffer, trace_ctx, 6, NULL);
3486 }
3487
3488 out:
3489 ring_buffer_nest_end(buffer);
3490 out_put:
3491 put_trace_buf();
3492
3493 out_nobuffer:
3494 preempt_enable_notrace();
3495 unpause_graph_tracing();
3496
3497 return len;
3498 }
3499 EXPORT_SYMBOL_GPL(trace_vbprintk);
3500
3501 __printf(3, 0)
3502 static int
__trace_array_vprintk(struct trace_buffer * buffer,unsigned long ip,const char * fmt,va_list args)3503 __trace_array_vprintk(struct trace_buffer *buffer,
3504 unsigned long ip, const char *fmt, va_list args)
3505 {
3506 struct trace_event_call *call = &event_print;
3507 struct ring_buffer_event *event;
3508 int len = 0, size;
3509 struct print_entry *entry;
3510 unsigned int trace_ctx;
3511 char *tbuffer;
3512
3513 if (tracing_disabled)
3514 return 0;
3515
3516 /* Don't pollute graph traces with trace_vprintk internals */
3517 pause_graph_tracing();
3518
3519 trace_ctx = tracing_gen_ctx();
3520 preempt_disable_notrace();
3521
3522
3523 tbuffer = get_trace_buf();
3524 if (!tbuffer) {
3525 len = 0;
3526 goto out_nobuffer;
3527 }
3528
3529 len = vscnprintf(tbuffer, TRACE_BUF_SIZE, fmt, args);
3530
3531 size = sizeof(*entry) + len + 1;
3532 ring_buffer_nest_start(buffer);
3533 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3534 trace_ctx);
3535 if (!event)
3536 goto out;
3537 entry = ring_buffer_event_data(event);
3538 entry->ip = ip;
3539
3540 memcpy(&entry->buf, tbuffer, len + 1);
3541 if (!call_filter_check_discard(call, entry, buffer, event)) {
3542 __buffer_unlock_commit(buffer, event);
3543 ftrace_trace_stack(&global_trace, buffer, trace_ctx, 6, NULL);
3544 }
3545
3546 out:
3547 ring_buffer_nest_end(buffer);
3548 put_trace_buf();
3549
3550 out_nobuffer:
3551 preempt_enable_notrace();
3552 unpause_graph_tracing();
3553
3554 return len;
3555 }
3556
3557 __printf(3, 0)
trace_array_vprintk(struct trace_array * tr,unsigned long ip,const char * fmt,va_list args)3558 int trace_array_vprintk(struct trace_array *tr,
3559 unsigned long ip, const char *fmt, va_list args)
3560 {
3561 if (tracing_selftest_running && tr == &global_trace)
3562 return 0;
3563
3564 return __trace_array_vprintk(tr->array_buffer.buffer, ip, fmt, args);
3565 }
3566
3567 /**
3568 * trace_array_printk - Print a message to a specific instance
3569 * @tr: The instance trace_array descriptor
3570 * @ip: The instruction pointer that this is called from.
3571 * @fmt: The format to print (printf format)
3572 *
3573 * If a subsystem sets up its own instance, they have the right to
3574 * printk strings into their tracing instance buffer using this
3575 * function. Note, this function will not write into the top level
3576 * buffer (use trace_printk() for that), as writing into the top level
3577 * buffer should only have events that can be individually disabled.
3578 * trace_printk() is only used for debugging a kernel, and should not
3579 * be ever incorporated in normal use.
3580 *
3581 * trace_array_printk() can be used, as it will not add noise to the
3582 * top level tracing buffer.
3583 *
3584 * Note, trace_array_init_printk() must be called on @tr before this
3585 * can be used.
3586 */
3587 __printf(3, 0)
trace_array_printk(struct trace_array * tr,unsigned long ip,const char * fmt,...)3588 int trace_array_printk(struct trace_array *tr,
3589 unsigned long ip, const char *fmt, ...)
3590 {
3591 int ret;
3592 va_list ap;
3593
3594 if (!tr)
3595 return -ENOENT;
3596
3597 /* This is only allowed for created instances */
3598 if (tr == &global_trace)
3599 return 0;
3600
3601 if (!(tr->trace_flags & TRACE_ITER_PRINTK))
3602 return 0;
3603
3604 va_start(ap, fmt);
3605 ret = trace_array_vprintk(tr, ip, fmt, ap);
3606 va_end(ap);
3607 return ret;
3608 }
3609 EXPORT_SYMBOL_GPL(trace_array_printk);
3610
3611 /**
3612 * trace_array_init_printk - Initialize buffers for trace_array_printk()
3613 * @tr: The trace array to initialize the buffers for
3614 *
3615 * As trace_array_printk() only writes into instances, they are OK to
3616 * have in the kernel (unlike trace_printk()). This needs to be called
3617 * before trace_array_printk() can be used on a trace_array.
3618 */
trace_array_init_printk(struct trace_array * tr)3619 int trace_array_init_printk(struct trace_array *tr)
3620 {
3621 if (!tr)
3622 return -ENOENT;
3623
3624 /* This is only allowed for created instances */
3625 if (tr == &global_trace)
3626 return -EINVAL;
3627
3628 return alloc_percpu_trace_buffer();
3629 }
3630 EXPORT_SYMBOL_GPL(trace_array_init_printk);
3631
3632 __printf(3, 4)
trace_array_printk_buf(struct trace_buffer * buffer,unsigned long ip,const char * fmt,...)3633 int trace_array_printk_buf(struct trace_buffer *buffer,
3634 unsigned long ip, const char *fmt, ...)
3635 {
3636 int ret;
3637 va_list ap;
3638
3639 if (!(global_trace.trace_flags & TRACE_ITER_PRINTK))
3640 return 0;
3641
3642 va_start(ap, fmt);
3643 ret = __trace_array_vprintk(buffer, ip, fmt, ap);
3644 va_end(ap);
3645 return ret;
3646 }
3647
3648 __printf(2, 0)
trace_vprintk(unsigned long ip,const char * fmt,va_list args)3649 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
3650 {
3651 return trace_array_vprintk(&global_trace, ip, fmt, args);
3652 }
3653 EXPORT_SYMBOL_GPL(trace_vprintk);
3654
trace_iterator_increment(struct trace_iterator * iter)3655 static void trace_iterator_increment(struct trace_iterator *iter)
3656 {
3657 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, iter->cpu);
3658
3659 iter->idx++;
3660 if (buf_iter)
3661 ring_buffer_iter_advance(buf_iter);
3662 }
3663
3664 static struct trace_entry *
peek_next_entry(struct trace_iterator * iter,int cpu,u64 * ts,unsigned long * lost_events)3665 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
3666 unsigned long *lost_events)
3667 {
3668 struct ring_buffer_event *event;
3669 struct ring_buffer_iter *buf_iter = trace_buffer_iter(iter, cpu);
3670
3671 if (buf_iter) {
3672 event = ring_buffer_iter_peek(buf_iter, ts);
3673 if (lost_events)
3674 *lost_events = ring_buffer_iter_dropped(buf_iter) ?
3675 (unsigned long)-1 : 0;
3676 } else {
3677 event = ring_buffer_peek(iter->array_buffer->buffer, cpu, ts,
3678 lost_events);
3679 }
3680
3681 if (event) {
3682 iter->ent_size = ring_buffer_event_length(event);
3683 return ring_buffer_event_data(event);
3684 }
3685 iter->ent_size = 0;
3686 return NULL;
3687 }
3688
3689 static struct trace_entry *
__find_next_entry(struct trace_iterator * iter,int * ent_cpu,unsigned long * missing_events,u64 * ent_ts)3690 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
3691 unsigned long *missing_events, u64 *ent_ts)
3692 {
3693 struct trace_buffer *buffer = iter->array_buffer->buffer;
3694 struct trace_entry *ent, *next = NULL;
3695 unsigned long lost_events = 0, next_lost = 0;
3696 int cpu_file = iter->cpu_file;
3697 u64 next_ts = 0, ts;
3698 int next_cpu = -1;
3699 int next_size = 0;
3700 int cpu;
3701
3702 /*
3703 * If we are in a per_cpu trace file, don't bother by iterating over
3704 * all cpu and peek directly.
3705 */
3706 if (cpu_file > RING_BUFFER_ALL_CPUS) {
3707 if (ring_buffer_empty_cpu(buffer, cpu_file))
3708 return NULL;
3709 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
3710 if (ent_cpu)
3711 *ent_cpu = cpu_file;
3712
3713 return ent;
3714 }
3715
3716 for_each_tracing_cpu(cpu) {
3717
3718 if (ring_buffer_empty_cpu(buffer, cpu))
3719 continue;
3720
3721 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
3722
3723 /*
3724 * Pick the entry with the smallest timestamp:
3725 */
3726 if (ent && (!next || ts < next_ts)) {
3727 next = ent;
3728 next_cpu = cpu;
3729 next_ts = ts;
3730 next_lost = lost_events;
3731 next_size = iter->ent_size;
3732 }
3733 }
3734
3735 iter->ent_size = next_size;
3736
3737 if (ent_cpu)
3738 *ent_cpu = next_cpu;
3739
3740 if (ent_ts)
3741 *ent_ts = next_ts;
3742
3743 if (missing_events)
3744 *missing_events = next_lost;
3745
3746 return next;
3747 }
3748
3749 #define STATIC_FMT_BUF_SIZE 128
3750 static char static_fmt_buf[STATIC_FMT_BUF_SIZE];
3751
trace_iter_expand_format(struct trace_iterator * iter)3752 char *trace_iter_expand_format(struct trace_iterator *iter)
3753 {
3754 char *tmp;
3755
3756 /*
3757 * iter->tr is NULL when used with tp_printk, which makes
3758 * this get called where it is not safe to call krealloc().
3759 */
3760 if (!iter->tr || iter->fmt == static_fmt_buf)
3761 return NULL;
3762
3763 tmp = krealloc(iter->fmt, iter->fmt_size + STATIC_FMT_BUF_SIZE,
3764 GFP_KERNEL);
3765 if (tmp) {
3766 iter->fmt_size += STATIC_FMT_BUF_SIZE;
3767 iter->fmt = tmp;
3768 }
3769
3770 return tmp;
3771 }
3772
3773 /* Returns true if the string is safe to dereference from an event */
trace_safe_str(struct trace_iterator * iter,const char * str,bool star,int len)3774 static bool trace_safe_str(struct trace_iterator *iter, const char *str,
3775 bool star, int len)
3776 {
3777 unsigned long addr = (unsigned long)str;
3778 struct trace_event *trace_event;
3779 struct trace_event_call *event;
3780
3781 /* Ignore strings with no length */
3782 if (star && !len)
3783 return true;
3784
3785 /* OK if part of the event data */
3786 if ((addr >= (unsigned long)iter->ent) &&
3787 (addr < (unsigned long)iter->ent + iter->ent_size))
3788 return true;
3789
3790 /* OK if part of the temp seq buffer */
3791 if ((addr >= (unsigned long)iter->tmp_seq.buffer) &&
3792 (addr < (unsigned long)iter->tmp_seq.buffer + TRACE_SEQ_BUFFER_SIZE))
3793 return true;
3794
3795 /* Core rodata can not be freed */
3796 if (is_kernel_rodata(addr))
3797 return true;
3798
3799 if (trace_is_tracepoint_string(str))
3800 return true;
3801
3802 /*
3803 * Now this could be a module event, referencing core module
3804 * data, which is OK.
3805 */
3806 if (!iter->ent)
3807 return false;
3808
3809 trace_event = ftrace_find_event(iter->ent->type);
3810 if (!trace_event)
3811 return false;
3812
3813 event = container_of(trace_event, struct trace_event_call, event);
3814 if ((event->flags & TRACE_EVENT_FL_DYNAMIC) || !event->module)
3815 return false;
3816
3817 /* Would rather have rodata, but this will suffice */
3818 if (within_module_core(addr, event->module))
3819 return true;
3820
3821 return false;
3822 }
3823
3824 static DEFINE_STATIC_KEY_FALSE(trace_no_verify);
3825
test_can_verify_check(const char * fmt,...)3826 static int test_can_verify_check(const char *fmt, ...)
3827 {
3828 char buf[16];
3829 va_list ap;
3830 int ret;
3831
3832 /*
3833 * The verifier is dependent on vsnprintf() modifies the va_list
3834 * passed to it, where it is sent as a reference. Some architectures
3835 * (like x86_32) passes it by value, which means that vsnprintf()
3836 * does not modify the va_list passed to it, and the verifier
3837 * would then need to be able to understand all the values that
3838 * vsnprintf can use. If it is passed by value, then the verifier
3839 * is disabled.
3840 */
3841 va_start(ap, fmt);
3842 vsnprintf(buf, 16, "%d", ap);
3843 ret = va_arg(ap, int);
3844 va_end(ap);
3845
3846 return ret;
3847 }
3848
test_can_verify(void)3849 static void test_can_verify(void)
3850 {
3851 if (!test_can_verify_check("%d %d", 0, 1)) {
3852 pr_info("trace event string verifier disabled\n");
3853 static_branch_inc(&trace_no_verify);
3854 }
3855 }
3856
3857 /**
3858 * trace_check_vprintf - Check dereferenced strings while writing to the seq buffer
3859 * @iter: The iterator that holds the seq buffer and the event being printed
3860 * @fmt: The format used to print the event
3861 * @ap: The va_list holding the data to print from @fmt.
3862 *
3863 * This writes the data into the @iter->seq buffer using the data from
3864 * @fmt and @ap. If the format has a %s, then the source of the string
3865 * is examined to make sure it is safe to print, otherwise it will
3866 * warn and print "[UNSAFE MEMORY]" in place of the dereferenced string
3867 * pointer.
3868 */
trace_check_vprintf(struct trace_iterator * iter,const char * fmt,va_list ap)3869 void trace_check_vprintf(struct trace_iterator *iter, const char *fmt,
3870 va_list ap)
3871 {
3872 const char *p = fmt;
3873 const char *str;
3874 int i, j;
3875
3876 if (WARN_ON_ONCE(!fmt))
3877 return;
3878
3879 if (static_branch_unlikely(&trace_no_verify))
3880 goto print;
3881
3882 /* Don't bother checking when doing a ftrace_dump() */
3883 if (iter->fmt == static_fmt_buf)
3884 goto print;
3885
3886 while (*p) {
3887 bool star = false;
3888 int len = 0;
3889
3890 j = 0;
3891
3892 /* We only care about %s and variants */
3893 for (i = 0; p[i]; i++) {
3894 if (i + 1 >= iter->fmt_size) {
3895 /*
3896 * If we can't expand the copy buffer,
3897 * just print it.
3898 */
3899 if (!trace_iter_expand_format(iter))
3900 goto print;
3901 }
3902
3903 if (p[i] == '\\' && p[i+1]) {
3904 i++;
3905 continue;
3906 }
3907 if (p[i] == '%') {
3908 /* Need to test cases like %08.*s */
3909 for (j = 1; p[i+j]; j++) {
3910 if (isdigit(p[i+j]) ||
3911 p[i+j] == '.')
3912 continue;
3913 if (p[i+j] == '*') {
3914 star = true;
3915 continue;
3916 }
3917 break;
3918 }
3919 if (p[i+j] == 's')
3920 break;
3921 star = false;
3922 }
3923 j = 0;
3924 }
3925 /* If no %s found then just print normally */
3926 if (!p[i])
3927 break;
3928
3929 /* Copy up to the %s, and print that */
3930 strncpy(iter->fmt, p, i);
3931 iter->fmt[i] = '\0';
3932 trace_seq_vprintf(&iter->seq, iter->fmt, ap);
3933
3934 /*
3935 * If iter->seq is full, the above call no longer guarantees
3936 * that ap is in sync with fmt processing, and further calls
3937 * to va_arg() can return wrong positional arguments.
3938 *
3939 * Ensure that ap is no longer used in this case.
3940 */
3941 if (iter->seq.full) {
3942 p = "";
3943 break;
3944 }
3945
3946 if (star)
3947 len = va_arg(ap, int);
3948
3949 /* The ap now points to the string data of the %s */
3950 str = va_arg(ap, const char *);
3951
3952 /*
3953 * If you hit this warning, it is likely that the
3954 * trace event in question used %s on a string that
3955 * was saved at the time of the event, but may not be
3956 * around when the trace is read. Use __string(),
3957 * __assign_str() and __get_str() helpers in the TRACE_EVENT()
3958 * instead. See samples/trace_events/trace-events-sample.h
3959 * for reference.
3960 */
3961 if (WARN_ONCE(!trace_safe_str(iter, str, star, len),
3962 "fmt: '%s' current_buffer: '%s'",
3963 fmt, seq_buf_str(&iter->seq.seq))) {
3964 int ret;
3965
3966 /* Try to safely read the string */
3967 if (star) {
3968 if (len + 1 > iter->fmt_size)
3969 len = iter->fmt_size - 1;
3970 if (len < 0)
3971 len = 0;
3972 ret = copy_from_kernel_nofault(iter->fmt, str, len);
3973 iter->fmt[len] = 0;
3974 star = false;
3975 } else {
3976 ret = strncpy_from_kernel_nofault(iter->fmt, str,
3977 iter->fmt_size);
3978 }
3979 if (ret < 0)
3980 trace_seq_printf(&iter->seq, "(0x%px)", str);
3981 else
3982 trace_seq_printf(&iter->seq, "(0x%px:%s)",
3983 str, iter->fmt);
3984 str = "[UNSAFE-MEMORY]";
3985 strcpy(iter->fmt, "%s");
3986 } else {
3987 strncpy(iter->fmt, p + i, j + 1);
3988 iter->fmt[j+1] = '\0';
3989 }
3990 if (star)
3991 trace_seq_printf(&iter->seq, iter->fmt, len, str);
3992 else
3993 trace_seq_printf(&iter->seq, iter->fmt, str);
3994
3995 p += i + j + 1;
3996 }
3997 print:
3998 if (*p)
3999 trace_seq_vprintf(&iter->seq, p, ap);
4000 }
4001
trace_event_format(struct trace_iterator * iter,const char * fmt)4002 const char *trace_event_format(struct trace_iterator *iter, const char *fmt)
4003 {
4004 const char *p, *new_fmt;
4005 char *q;
4006
4007 if (WARN_ON_ONCE(!fmt))
4008 return fmt;
4009
4010 if (!iter->tr || iter->tr->trace_flags & TRACE_ITER_HASH_PTR)
4011 return fmt;
4012
4013 p = fmt;
4014 new_fmt = q = iter->fmt;
4015 while (*p) {
4016 if (unlikely(q - new_fmt + 3 > iter->fmt_size)) {
4017 if (!trace_iter_expand_format(iter))
4018 return fmt;
4019
4020 q += iter->fmt - new_fmt;
4021 new_fmt = iter->fmt;
4022 }
4023
4024 *q++ = *p++;
4025
4026 /* Replace %p with %px */
4027 if (p[-1] == '%') {
4028 if (p[0] == '%') {
4029 *q++ = *p++;
4030 } else if (p[0] == 'p' && !isalnum(p[1])) {
4031 *q++ = *p++;
4032 *q++ = 'x';
4033 }
4034 }
4035 }
4036 *q = '\0';
4037
4038 return new_fmt;
4039 }
4040
4041 #define STATIC_TEMP_BUF_SIZE 128
4042 static char static_temp_buf[STATIC_TEMP_BUF_SIZE] __aligned(4);
4043
4044 /* Find the next real entry, without updating the iterator itself */
trace_find_next_entry(struct trace_iterator * iter,int * ent_cpu,u64 * ent_ts)4045 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
4046 int *ent_cpu, u64 *ent_ts)
4047 {
4048 /* __find_next_entry will reset ent_size */
4049 int ent_size = iter->ent_size;
4050 struct trace_entry *entry;
4051
4052 /*
4053 * If called from ftrace_dump(), then the iter->temp buffer
4054 * will be the static_temp_buf and not created from kmalloc.
4055 * If the entry size is greater than the buffer, we can
4056 * not save it. Just return NULL in that case. This is only
4057 * used to add markers when two consecutive events' time
4058 * stamps have a large delta. See trace_print_lat_context()
4059 */
4060 if (iter->temp == static_temp_buf &&
4061 STATIC_TEMP_BUF_SIZE < ent_size)
4062 return NULL;
4063
4064 /*
4065 * The __find_next_entry() may call peek_next_entry(), which may
4066 * call ring_buffer_peek() that may make the contents of iter->ent
4067 * undefined. Need to copy iter->ent now.
4068 */
4069 if (iter->ent && iter->ent != iter->temp) {
4070 if ((!iter->temp || iter->temp_size < iter->ent_size) &&
4071 !WARN_ON_ONCE(iter->temp == static_temp_buf)) {
4072 void *temp;
4073 temp = kmalloc(iter->ent_size, GFP_KERNEL);
4074 if (!temp)
4075 return NULL;
4076 kfree(iter->temp);
4077 iter->temp = temp;
4078 iter->temp_size = iter->ent_size;
4079 }
4080 memcpy(iter->temp, iter->ent, iter->ent_size);
4081 iter->ent = iter->temp;
4082 }
4083 entry = __find_next_entry(iter, ent_cpu, NULL, ent_ts);
4084 /* Put back the original ent_size */
4085 iter->ent_size = ent_size;
4086
4087 return entry;
4088 }
4089
4090 /* Find the next real entry, and increment the iterator to the next entry */
trace_find_next_entry_inc(struct trace_iterator * iter)4091 void *trace_find_next_entry_inc(struct trace_iterator *iter)
4092 {
4093 iter->ent = __find_next_entry(iter, &iter->cpu,
4094 &iter->lost_events, &iter->ts);
4095
4096 if (iter->ent)
4097 trace_iterator_increment(iter);
4098
4099 return iter->ent ? iter : NULL;
4100 }
4101
trace_consume(struct trace_iterator * iter)4102 static void trace_consume(struct trace_iterator *iter)
4103 {
4104 ring_buffer_consume(iter->array_buffer->buffer, iter->cpu, &iter->ts,
4105 &iter->lost_events);
4106 }
4107
s_next(struct seq_file * m,void * v,loff_t * pos)4108 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
4109 {
4110 struct trace_iterator *iter = m->private;
4111 int i = (int)*pos;
4112 void *ent;
4113
4114 WARN_ON_ONCE(iter->leftover);
4115
4116 (*pos)++;
4117
4118 /* can't go backwards */
4119 if (iter->idx > i)
4120 return NULL;
4121
4122 if (iter->idx < 0)
4123 ent = trace_find_next_entry_inc(iter);
4124 else
4125 ent = iter;
4126
4127 while (ent && iter->idx < i)
4128 ent = trace_find_next_entry_inc(iter);
4129
4130 iter->pos = *pos;
4131
4132 return ent;
4133 }
4134
tracing_iter_reset(struct trace_iterator * iter,int cpu)4135 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
4136 {
4137 struct ring_buffer_iter *buf_iter;
4138 unsigned long entries = 0;
4139 u64 ts;
4140
4141 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = 0;
4142
4143 buf_iter = trace_buffer_iter(iter, cpu);
4144 if (!buf_iter)
4145 return;
4146
4147 ring_buffer_iter_reset(buf_iter);
4148
4149 /*
4150 * We could have the case with the max latency tracers
4151 * that a reset never took place on a cpu. This is evident
4152 * by the timestamp being before the start of the buffer.
4153 */
4154 while (ring_buffer_iter_peek(buf_iter, &ts)) {
4155 if (ts >= iter->array_buffer->time_start)
4156 break;
4157 entries++;
4158 ring_buffer_iter_advance(buf_iter);
4159 }
4160
4161 per_cpu_ptr(iter->array_buffer->data, cpu)->skipped_entries = entries;
4162 }
4163
4164 /*
4165 * The current tracer is copied to avoid a global locking
4166 * all around.
4167 */
s_start(struct seq_file * m,loff_t * pos)4168 static void *s_start(struct seq_file *m, loff_t *pos)
4169 {
4170 struct trace_iterator *iter = m->private;
4171 struct trace_array *tr = iter->tr;
4172 int cpu_file = iter->cpu_file;
4173 void *p = NULL;
4174 loff_t l = 0;
4175 int cpu;
4176
4177 mutex_lock(&trace_types_lock);
4178 if (unlikely(tr->current_trace != iter->trace)) {
4179 /* Close iter->trace before switching to the new current tracer */
4180 if (iter->trace->close)
4181 iter->trace->close(iter);
4182 iter->trace = tr->current_trace;
4183 /* Reopen the new current tracer */
4184 if (iter->trace->open)
4185 iter->trace->open(iter);
4186 }
4187 mutex_unlock(&trace_types_lock);
4188
4189 #ifdef CONFIG_TRACER_MAX_TRACE
4190 if (iter->snapshot && iter->trace->use_max_tr)
4191 return ERR_PTR(-EBUSY);
4192 #endif
4193
4194 if (*pos != iter->pos) {
4195 iter->ent = NULL;
4196 iter->cpu = 0;
4197 iter->idx = -1;
4198
4199 if (cpu_file == RING_BUFFER_ALL_CPUS) {
4200 for_each_tracing_cpu(cpu)
4201 tracing_iter_reset(iter, cpu);
4202 } else
4203 tracing_iter_reset(iter, cpu_file);
4204
4205 iter->leftover = 0;
4206 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
4207 ;
4208
4209 } else {
4210 /*
4211 * If we overflowed the seq_file before, then we want
4212 * to just reuse the trace_seq buffer again.
4213 */
4214 if (iter->leftover)
4215 p = iter;
4216 else {
4217 l = *pos - 1;
4218 p = s_next(m, p, &l);
4219 }
4220 }
4221
4222 trace_event_read_lock();
4223 trace_access_lock(cpu_file);
4224 return p;
4225 }
4226
s_stop(struct seq_file * m,void * p)4227 static void s_stop(struct seq_file *m, void *p)
4228 {
4229 struct trace_iterator *iter = m->private;
4230
4231 #ifdef CONFIG_TRACER_MAX_TRACE
4232 if (iter->snapshot && iter->trace->use_max_tr)
4233 return;
4234 #endif
4235
4236 trace_access_unlock(iter->cpu_file);
4237 trace_event_read_unlock();
4238 }
4239
4240 static void
get_total_entries_cpu(struct array_buffer * buf,unsigned long * total,unsigned long * entries,int cpu)4241 get_total_entries_cpu(struct array_buffer *buf, unsigned long *total,
4242 unsigned long *entries, int cpu)
4243 {
4244 unsigned long count;
4245
4246 count = ring_buffer_entries_cpu(buf->buffer, cpu);
4247 /*
4248 * If this buffer has skipped entries, then we hold all
4249 * entries for the trace and we need to ignore the
4250 * ones before the time stamp.
4251 */
4252 if (per_cpu_ptr(buf->data, cpu)->skipped_entries) {
4253 count -= per_cpu_ptr(buf->data, cpu)->skipped_entries;
4254 /* total is the same as the entries */
4255 *total = count;
4256 } else
4257 *total = count +
4258 ring_buffer_overrun_cpu(buf->buffer, cpu);
4259 *entries = count;
4260 }
4261
4262 static void
get_total_entries(struct array_buffer * buf,unsigned long * total,unsigned long * entries)4263 get_total_entries(struct array_buffer *buf,
4264 unsigned long *total, unsigned long *entries)
4265 {
4266 unsigned long t, e;
4267 int cpu;
4268
4269 *total = 0;
4270 *entries = 0;
4271
4272 for_each_tracing_cpu(cpu) {
4273 get_total_entries_cpu(buf, &t, &e, cpu);
4274 *total += t;
4275 *entries += e;
4276 }
4277 }
4278
trace_total_entries_cpu(struct trace_array * tr,int cpu)4279 unsigned long trace_total_entries_cpu(struct trace_array *tr, int cpu)
4280 {
4281 unsigned long total, entries;
4282
4283 if (!tr)
4284 tr = &global_trace;
4285
4286 get_total_entries_cpu(&tr->array_buffer, &total, &entries, cpu);
4287
4288 return entries;
4289 }
4290
trace_total_entries(struct trace_array * tr)4291 unsigned long trace_total_entries(struct trace_array *tr)
4292 {
4293 unsigned long total, entries;
4294
4295 if (!tr)
4296 tr = &global_trace;
4297
4298 get_total_entries(&tr->array_buffer, &total, &entries);
4299
4300 return entries;
4301 }
4302
print_lat_help_header(struct seq_file * m)4303 static void print_lat_help_header(struct seq_file *m)
4304 {
4305 seq_puts(m, "# _------=> CPU# \n"
4306 "# / _-----=> irqs-off/BH-disabled\n"
4307 "# | / _----=> need-resched \n"
4308 "# || / _---=> hardirq/softirq \n"
4309 "# ||| / _--=> preempt-depth \n"
4310 "# |||| / _-=> migrate-disable \n"
4311 "# ||||| / delay \n"
4312 "# cmd pid |||||| time | caller \n"
4313 "# \\ / |||||| \\ | / \n");
4314 }
4315
print_event_info(struct array_buffer * buf,struct seq_file * m)4316 static void print_event_info(struct array_buffer *buf, struct seq_file *m)
4317 {
4318 unsigned long total;
4319 unsigned long entries;
4320
4321 get_total_entries(buf, &total, &entries);
4322 seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu #P:%d\n",
4323 entries, total, num_online_cpus());
4324 seq_puts(m, "#\n");
4325 }
4326
print_func_help_header(struct array_buffer * buf,struct seq_file * m,unsigned int flags)4327 static void print_func_help_header(struct array_buffer *buf, struct seq_file *m,
4328 unsigned int flags)
4329 {
4330 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4331
4332 print_event_info(buf, m);
4333
4334 seq_printf(m, "# TASK-PID %s CPU# TIMESTAMP FUNCTION\n", tgid ? " TGID " : "");
4335 seq_printf(m, "# | | %s | | |\n", tgid ? " | " : "");
4336 }
4337
print_func_help_header_irq(struct array_buffer * buf,struct seq_file * m,unsigned int flags)4338 static void print_func_help_header_irq(struct array_buffer *buf, struct seq_file *m,
4339 unsigned int flags)
4340 {
4341 bool tgid = flags & TRACE_ITER_RECORD_TGID;
4342 static const char space[] = " ";
4343 int prec = tgid ? 12 : 2;
4344
4345 print_event_info(buf, m);
4346
4347 seq_printf(m, "# %.*s _-----=> irqs-off/BH-disabled\n", prec, space);
4348 seq_printf(m, "# %.*s / _----=> need-resched\n", prec, space);
4349 seq_printf(m, "# %.*s| / _---=> hardirq/softirq\n", prec, space);
4350 seq_printf(m, "# %.*s|| / _--=> preempt-depth\n", prec, space);
4351 seq_printf(m, "# %.*s||| / _-=> migrate-disable\n", prec, space);
4352 seq_printf(m, "# %.*s|||| / delay\n", prec, space);
4353 seq_printf(m, "# TASK-PID %.*s CPU# ||||| TIMESTAMP FUNCTION\n", prec, " TGID ");
4354 seq_printf(m, "# | | %.*s | ||||| | |\n", prec, " | ");
4355 }
4356
4357 void
print_trace_header(struct seq_file * m,struct trace_iterator * iter)4358 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
4359 {
4360 unsigned long sym_flags = (global_trace.trace_flags & TRACE_ITER_SYM_MASK);
4361 struct array_buffer *buf = iter->array_buffer;
4362 struct trace_array_cpu *data = per_cpu_ptr(buf->data, buf->cpu);
4363 struct tracer *type = iter->trace;
4364 unsigned long entries;
4365 unsigned long total;
4366 const char *name = type->name;
4367
4368 get_total_entries(buf, &total, &entries);
4369
4370 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
4371 name, UTS_RELEASE);
4372 seq_puts(m, "# -----------------------------------"
4373 "---------------------------------\n");
4374 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
4375 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
4376 nsecs_to_usecs(data->saved_latency),
4377 entries,
4378 total,
4379 buf->cpu,
4380 preempt_model_none() ? "server" :
4381 preempt_model_voluntary() ? "desktop" :
4382 preempt_model_full() ? "preempt" :
4383 preempt_model_rt() ? "preempt_rt" :
4384 "unknown",
4385 /* These are reserved for later use */
4386 0, 0, 0, 0);
4387 #ifdef CONFIG_SMP
4388 seq_printf(m, " #P:%d)\n", num_online_cpus());
4389 #else
4390 seq_puts(m, ")\n");
4391 #endif
4392 seq_puts(m, "# -----------------\n");
4393 seq_printf(m, "# | task: %.16s-%d "
4394 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
4395 data->comm, data->pid,
4396 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
4397 data->policy, data->rt_priority);
4398 seq_puts(m, "# -----------------\n");
4399
4400 if (data->critical_start) {
4401 seq_puts(m, "# => started at: ");
4402 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
4403 trace_print_seq(m, &iter->seq);
4404 seq_puts(m, "\n# => ended at: ");
4405 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
4406 trace_print_seq(m, &iter->seq);
4407 seq_puts(m, "\n#\n");
4408 }
4409
4410 seq_puts(m, "#\n");
4411 }
4412
test_cpu_buff_start(struct trace_iterator * iter)4413 static void test_cpu_buff_start(struct trace_iterator *iter)
4414 {
4415 struct trace_seq *s = &iter->seq;
4416 struct trace_array *tr = iter->tr;
4417
4418 if (!(tr->trace_flags & TRACE_ITER_ANNOTATE))
4419 return;
4420
4421 if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
4422 return;
4423
4424 if (cpumask_available(iter->started) &&
4425 cpumask_test_cpu(iter->cpu, iter->started))
4426 return;
4427
4428 if (per_cpu_ptr(iter->array_buffer->data, iter->cpu)->skipped_entries)
4429 return;
4430
4431 if (cpumask_available(iter->started))
4432 cpumask_set_cpu(iter->cpu, iter->started);
4433
4434 /* Don't print started cpu buffer for the first entry of the trace */
4435 if (iter->idx > 1)
4436 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
4437 iter->cpu);
4438 }
4439
print_trace_fmt(struct trace_iterator * iter)4440 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
4441 {
4442 struct trace_array *tr = iter->tr;
4443 struct trace_seq *s = &iter->seq;
4444 unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
4445 struct trace_entry *entry;
4446 struct trace_event *event;
4447
4448 entry = iter->ent;
4449
4450 test_cpu_buff_start(iter);
4451
4452 event = ftrace_find_event(entry->type);
4453
4454 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4455 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4456 trace_print_lat_context(iter);
4457 else
4458 trace_print_context(iter);
4459 }
4460
4461 if (trace_seq_has_overflowed(s))
4462 return TRACE_TYPE_PARTIAL_LINE;
4463
4464 if (event) {
4465 if (tr->trace_flags & TRACE_ITER_FIELDS)
4466 return print_event_fields(iter, event);
4467 return event->funcs->trace(iter, sym_flags, event);
4468 }
4469
4470 trace_seq_printf(s, "Unknown type %d\n", entry->type);
4471
4472 return trace_handle_return(s);
4473 }
4474
print_raw_fmt(struct trace_iterator * iter)4475 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
4476 {
4477 struct trace_array *tr = iter->tr;
4478 struct trace_seq *s = &iter->seq;
4479 struct trace_entry *entry;
4480 struct trace_event *event;
4481
4482 entry = iter->ent;
4483
4484 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO)
4485 trace_seq_printf(s, "%d %d %llu ",
4486 entry->pid, iter->cpu, iter->ts);
4487
4488 if (trace_seq_has_overflowed(s))
4489 return TRACE_TYPE_PARTIAL_LINE;
4490
4491 event = ftrace_find_event(entry->type);
4492 if (event)
4493 return event->funcs->raw(iter, 0, event);
4494
4495 trace_seq_printf(s, "%d ?\n", entry->type);
4496
4497 return trace_handle_return(s);
4498 }
4499
print_hex_fmt(struct trace_iterator * iter)4500 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
4501 {
4502 struct trace_array *tr = iter->tr;
4503 struct trace_seq *s = &iter->seq;
4504 unsigned char newline = '\n';
4505 struct trace_entry *entry;
4506 struct trace_event *event;
4507
4508 entry = iter->ent;
4509
4510 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4511 SEQ_PUT_HEX_FIELD(s, entry->pid);
4512 SEQ_PUT_HEX_FIELD(s, iter->cpu);
4513 SEQ_PUT_HEX_FIELD(s, iter->ts);
4514 if (trace_seq_has_overflowed(s))
4515 return TRACE_TYPE_PARTIAL_LINE;
4516 }
4517
4518 event = ftrace_find_event(entry->type);
4519 if (event) {
4520 enum print_line_t ret = event->funcs->hex(iter, 0, event);
4521 if (ret != TRACE_TYPE_HANDLED)
4522 return ret;
4523 }
4524
4525 SEQ_PUT_FIELD(s, newline);
4526
4527 return trace_handle_return(s);
4528 }
4529
print_bin_fmt(struct trace_iterator * iter)4530 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
4531 {
4532 struct trace_array *tr = iter->tr;
4533 struct trace_seq *s = &iter->seq;
4534 struct trace_entry *entry;
4535 struct trace_event *event;
4536
4537 entry = iter->ent;
4538
4539 if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
4540 SEQ_PUT_FIELD(s, entry->pid);
4541 SEQ_PUT_FIELD(s, iter->cpu);
4542 SEQ_PUT_FIELD(s, iter->ts);
4543 if (trace_seq_has_overflowed(s))
4544 return TRACE_TYPE_PARTIAL_LINE;
4545 }
4546
4547 event = ftrace_find_event(entry->type);
4548 return event ? event->funcs->binary(iter, 0, event) :
4549 TRACE_TYPE_HANDLED;
4550 }
4551
trace_empty(struct trace_iterator * iter)4552 int trace_empty(struct trace_iterator *iter)
4553 {
4554 struct ring_buffer_iter *buf_iter;
4555 int cpu;
4556
4557 /* If we are looking at one CPU buffer, only check that one */
4558 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
4559 cpu = iter->cpu_file;
4560 buf_iter = trace_buffer_iter(iter, cpu);
4561 if (buf_iter) {
4562 if (!ring_buffer_iter_empty(buf_iter))
4563 return 0;
4564 } else {
4565 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4566 return 0;
4567 }
4568 return 1;
4569 }
4570
4571 for_each_tracing_cpu(cpu) {
4572 buf_iter = trace_buffer_iter(iter, cpu);
4573 if (buf_iter) {
4574 if (!ring_buffer_iter_empty(buf_iter))
4575 return 0;
4576 } else {
4577 if (!ring_buffer_empty_cpu(iter->array_buffer->buffer, cpu))
4578 return 0;
4579 }
4580 }
4581
4582 return 1;
4583 }
4584
4585 /* Called with trace_event_read_lock() held. */
print_trace_line(struct trace_iterator * iter)4586 enum print_line_t print_trace_line(struct trace_iterator *iter)
4587 {
4588 struct trace_array *tr = iter->tr;
4589 unsigned long trace_flags = tr->trace_flags;
4590 enum print_line_t ret;
4591
4592 if (iter->lost_events) {
4593 if (iter->lost_events == (unsigned long)-1)
4594 trace_seq_printf(&iter->seq, "CPU:%d [LOST EVENTS]\n",
4595 iter->cpu);
4596 else
4597 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
4598 iter->cpu, iter->lost_events);
4599 if (trace_seq_has_overflowed(&iter->seq))
4600 return TRACE_TYPE_PARTIAL_LINE;
4601 }
4602
4603 if (iter->trace && iter->trace->print_line) {
4604 ret = iter->trace->print_line(iter);
4605 if (ret != TRACE_TYPE_UNHANDLED)
4606 return ret;
4607 }
4608
4609 if (iter->ent->type == TRACE_BPUTS &&
4610 trace_flags & TRACE_ITER_PRINTK &&
4611 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4612 return trace_print_bputs_msg_only(iter);
4613
4614 if (iter->ent->type == TRACE_BPRINT &&
4615 trace_flags & TRACE_ITER_PRINTK &&
4616 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4617 return trace_print_bprintk_msg_only(iter);
4618
4619 if (iter->ent->type == TRACE_PRINT &&
4620 trace_flags & TRACE_ITER_PRINTK &&
4621 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
4622 return trace_print_printk_msg_only(iter);
4623
4624 if (trace_flags & TRACE_ITER_BIN)
4625 return print_bin_fmt(iter);
4626
4627 if (trace_flags & TRACE_ITER_HEX)
4628 return print_hex_fmt(iter);
4629
4630 if (trace_flags & TRACE_ITER_RAW)
4631 return print_raw_fmt(iter);
4632
4633 return print_trace_fmt(iter);
4634 }
4635
trace_latency_header(struct seq_file * m)4636 void trace_latency_header(struct seq_file *m)
4637 {
4638 struct trace_iterator *iter = m->private;
4639 struct trace_array *tr = iter->tr;
4640
4641 /* print nothing if the buffers are empty */
4642 if (trace_empty(iter))
4643 return;
4644
4645 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
4646 print_trace_header(m, iter);
4647
4648 if (!(tr->trace_flags & TRACE_ITER_VERBOSE))
4649 print_lat_help_header(m);
4650 }
4651
trace_default_header(struct seq_file * m)4652 void trace_default_header(struct seq_file *m)
4653 {
4654 struct trace_iterator *iter = m->private;
4655 struct trace_array *tr = iter->tr;
4656 unsigned long trace_flags = tr->trace_flags;
4657
4658 if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
4659 return;
4660
4661 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
4662 /* print nothing if the buffers are empty */
4663 if (trace_empty(iter))
4664 return;
4665 print_trace_header(m, iter);
4666 if (!(trace_flags & TRACE_ITER_VERBOSE))
4667 print_lat_help_header(m);
4668 } else {
4669 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
4670 if (trace_flags & TRACE_ITER_IRQ_INFO)
4671 print_func_help_header_irq(iter->array_buffer,
4672 m, trace_flags);
4673 else
4674 print_func_help_header(iter->array_buffer, m,
4675 trace_flags);
4676 }
4677 }
4678 }
4679
test_ftrace_alive(struct seq_file * m)4680 static void test_ftrace_alive(struct seq_file *m)
4681 {
4682 if (!ftrace_is_dead())
4683 return;
4684 seq_puts(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n"
4685 "# MAY BE MISSING FUNCTION EVENTS\n");
4686 }
4687
4688 #ifdef CONFIG_TRACER_MAX_TRACE
show_snapshot_main_help(struct seq_file * m)4689 static void show_snapshot_main_help(struct seq_file *m)
4690 {
4691 seq_puts(m, "# echo 0 > snapshot : Clears and frees snapshot buffer\n"
4692 "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4693 "# Takes a snapshot of the main buffer.\n"
4694 "# echo 2 > snapshot : Clears snapshot buffer (but does not allocate or free)\n"
4695 "# (Doesn't have to be '2' works with any number that\n"
4696 "# is not a '0' or '1')\n");
4697 }
4698
show_snapshot_percpu_help(struct seq_file * m)4699 static void show_snapshot_percpu_help(struct seq_file *m)
4700 {
4701 seq_puts(m, "# echo 0 > snapshot : Invalid for per_cpu snapshot file.\n");
4702 #ifdef CONFIG_RING_BUFFER_ALLOW_SWAP
4703 seq_puts(m, "# echo 1 > snapshot : Allocates snapshot buffer, if not already allocated.\n"
4704 "# Takes a snapshot of the main buffer for this cpu.\n");
4705 #else
4706 seq_puts(m, "# echo 1 > snapshot : Not supported with this kernel.\n"
4707 "# Must use main snapshot file to allocate.\n");
4708 #endif
4709 seq_puts(m, "# echo 2 > snapshot : Clears this cpu's snapshot buffer (but does not allocate)\n"
4710 "# (Doesn't have to be '2' works with any number that\n"
4711 "# is not a '0' or '1')\n");
4712 }
4713
print_snapshot_help(struct seq_file * m,struct trace_iterator * iter)4714 static void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter)
4715 {
4716 if (iter->tr->allocated_snapshot)
4717 seq_puts(m, "#\n# * Snapshot is allocated *\n#\n");
4718 else
4719 seq_puts(m, "#\n# * Snapshot is freed *\n#\n");
4720
4721 seq_puts(m, "# Snapshot commands:\n");
4722 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
4723 show_snapshot_main_help(m);
4724 else
4725 show_snapshot_percpu_help(m);
4726 }
4727 #else
4728 /* Should never be called */
print_snapshot_help(struct seq_file * m,struct trace_iterator * iter)4729 static inline void print_snapshot_help(struct seq_file *m, struct trace_iterator *iter) { }
4730 #endif
4731
s_show(struct seq_file * m,void * v)4732 static int s_show(struct seq_file *m, void *v)
4733 {
4734 struct trace_iterator *iter = v;
4735 int ret;
4736
4737 if (iter->ent == NULL) {
4738 if (iter->tr) {
4739 seq_printf(m, "# tracer: %s\n", iter->trace->name);
4740 seq_puts(m, "#\n");
4741 test_ftrace_alive(m);
4742 }
4743 if (iter->snapshot && trace_empty(iter))
4744 print_snapshot_help(m, iter);
4745 else if (iter->trace && iter->trace->print_header)
4746 iter->trace->print_header(m);
4747 else
4748 trace_default_header(m);
4749
4750 } else if (iter->leftover) {
4751 /*
4752 * If we filled the seq_file buffer earlier, we
4753 * want to just show it now.
4754 */
4755 ret = trace_print_seq(m, &iter->seq);
4756
4757 /* ret should this time be zero, but you never know */
4758 iter->leftover = ret;
4759
4760 } else {
4761 ret = print_trace_line(iter);
4762 if (ret == TRACE_TYPE_PARTIAL_LINE) {
4763 iter->seq.full = 0;
4764 trace_seq_puts(&iter->seq, "[LINE TOO BIG]\n");
4765 }
4766 ret = trace_print_seq(m, &iter->seq);
4767 /*
4768 * If we overflow the seq_file buffer, then it will
4769 * ask us for this data again at start up.
4770 * Use that instead.
4771 * ret is 0 if seq_file write succeeded.
4772 * -1 otherwise.
4773 */
4774 iter->leftover = ret;
4775 }
4776
4777 return 0;
4778 }
4779
4780 /*
4781 * Should be used after trace_array_get(), trace_types_lock
4782 * ensures that i_cdev was already initialized.
4783 */
tracing_get_cpu(struct inode * inode)4784 static inline int tracing_get_cpu(struct inode *inode)
4785 {
4786 if (inode->i_cdev) /* See trace_create_cpu_file() */
4787 return (long)inode->i_cdev - 1;
4788 return RING_BUFFER_ALL_CPUS;
4789 }
4790
4791 static const struct seq_operations tracer_seq_ops = {
4792 .start = s_start,
4793 .next = s_next,
4794 .stop = s_stop,
4795 .show = s_show,
4796 };
4797
4798 /*
4799 * Note, as iter itself can be allocated and freed in different
4800 * ways, this function is only used to free its content, and not
4801 * the iterator itself. The only requirement to all the allocations
4802 * is that it must zero all fields (kzalloc), as freeing works with
4803 * ethier allocated content or NULL.
4804 */
free_trace_iter_content(struct trace_iterator * iter)4805 static void free_trace_iter_content(struct trace_iterator *iter)
4806 {
4807 /* The fmt is either NULL, allocated or points to static_fmt_buf */
4808 if (iter->fmt != static_fmt_buf)
4809 kfree(iter->fmt);
4810
4811 kfree(iter->temp);
4812 kfree(iter->buffer_iter);
4813 mutex_destroy(&iter->mutex);
4814 free_cpumask_var(iter->started);
4815 }
4816
4817 static struct trace_iterator *
__tracing_open(struct inode * inode,struct file * file,bool snapshot)4818 __tracing_open(struct inode *inode, struct file *file, bool snapshot)
4819 {
4820 struct trace_array *tr = inode->i_private;
4821 struct trace_iterator *iter;
4822 int cpu;
4823
4824 if (tracing_disabled)
4825 return ERR_PTR(-ENODEV);
4826
4827 iter = __seq_open_private(file, &tracer_seq_ops, sizeof(*iter));
4828 if (!iter)
4829 return ERR_PTR(-ENOMEM);
4830
4831 iter->buffer_iter = kcalloc(nr_cpu_ids, sizeof(*iter->buffer_iter),
4832 GFP_KERNEL);
4833 if (!iter->buffer_iter)
4834 goto release;
4835
4836 /*
4837 * trace_find_next_entry() may need to save off iter->ent.
4838 * It will place it into the iter->temp buffer. As most
4839 * events are less than 128, allocate a buffer of that size.
4840 * If one is greater, then trace_find_next_entry() will
4841 * allocate a new buffer to adjust for the bigger iter->ent.
4842 * It's not critical if it fails to get allocated here.
4843 */
4844 iter->temp = kmalloc(128, GFP_KERNEL);
4845 if (iter->temp)
4846 iter->temp_size = 128;
4847
4848 /*
4849 * trace_event_printf() may need to modify given format
4850 * string to replace %p with %px so that it shows real address
4851 * instead of hash value. However, that is only for the event
4852 * tracing, other tracer may not need. Defer the allocation
4853 * until it is needed.
4854 */
4855 iter->fmt = NULL;
4856 iter->fmt_size = 0;
4857
4858 mutex_lock(&trace_types_lock);
4859 iter->trace = tr->current_trace;
4860
4861 if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
4862 goto fail;
4863
4864 iter->tr = tr;
4865
4866 #ifdef CONFIG_TRACER_MAX_TRACE
4867 /* Currently only the top directory has a snapshot */
4868 if (tr->current_trace->print_max || snapshot)
4869 iter->array_buffer = &tr->max_buffer;
4870 else
4871 #endif
4872 iter->array_buffer = &tr->array_buffer;
4873 iter->snapshot = snapshot;
4874 iter->pos = -1;
4875 iter->cpu_file = tracing_get_cpu(inode);
4876 mutex_init(&iter->mutex);
4877
4878 /* Notify the tracer early; before we stop tracing. */
4879 if (iter->trace->open)
4880 iter->trace->open(iter);
4881
4882 /* Annotate start of buffers if we had overruns */
4883 if (ring_buffer_overruns(iter->array_buffer->buffer))
4884 iter->iter_flags |= TRACE_FILE_ANNOTATE;
4885
4886 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
4887 if (trace_clocks[tr->clock_id].in_ns)
4888 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
4889
4890 /*
4891 * If pause-on-trace is enabled, then stop the trace while
4892 * dumping, unless this is the "snapshot" file
4893 */
4894 if (!iter->snapshot && (tr->trace_flags & TRACE_ITER_PAUSE_ON_TRACE))
4895 tracing_stop_tr(tr);
4896
4897 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
4898 for_each_tracing_cpu(cpu) {
4899 iter->buffer_iter[cpu] =
4900 ring_buffer_read_prepare(iter->array_buffer->buffer,
4901 cpu, GFP_KERNEL);
4902 }
4903 ring_buffer_read_prepare_sync();
4904 for_each_tracing_cpu(cpu) {
4905 ring_buffer_read_start(iter->buffer_iter[cpu]);
4906 tracing_iter_reset(iter, cpu);
4907 }
4908 } else {
4909 cpu = iter->cpu_file;
4910 iter->buffer_iter[cpu] =
4911 ring_buffer_read_prepare(iter->array_buffer->buffer,
4912 cpu, GFP_KERNEL);
4913 ring_buffer_read_prepare_sync();
4914 ring_buffer_read_start(iter->buffer_iter[cpu]);
4915 tracing_iter_reset(iter, cpu);
4916 }
4917
4918 mutex_unlock(&trace_types_lock);
4919
4920 return iter;
4921
4922 fail:
4923 mutex_unlock(&trace_types_lock);
4924 free_trace_iter_content(iter);
4925 release:
4926 seq_release_private(inode, file);
4927 return ERR_PTR(-ENOMEM);
4928 }
4929
tracing_open_generic(struct inode * inode,struct file * filp)4930 int tracing_open_generic(struct inode *inode, struct file *filp)
4931 {
4932 int ret;
4933
4934 ret = tracing_check_open_get_tr(NULL);
4935 if (ret)
4936 return ret;
4937
4938 filp->private_data = inode->i_private;
4939 return 0;
4940 }
4941
tracing_is_disabled(void)4942 bool tracing_is_disabled(void)
4943 {
4944 return (tracing_disabled) ? true: false;
4945 }
4946
4947 /*
4948 * Open and update trace_array ref count.
4949 * Must have the current trace_array passed to it.
4950 */
tracing_open_generic_tr(struct inode * inode,struct file * filp)4951 int tracing_open_generic_tr(struct inode *inode, struct file *filp)
4952 {
4953 struct trace_array *tr = inode->i_private;
4954 int ret;
4955
4956 ret = tracing_check_open_get_tr(tr);
4957 if (ret)
4958 return ret;
4959
4960 filp->private_data = inode->i_private;
4961
4962 return 0;
4963 }
4964
4965 /*
4966 * The private pointer of the inode is the trace_event_file.
4967 * Update the tr ref count associated to it.
4968 */
tracing_open_file_tr(struct inode * inode,struct file * filp)4969 int tracing_open_file_tr(struct inode *inode, struct file *filp)
4970 {
4971 struct trace_event_file *file = inode->i_private;
4972 int ret;
4973
4974 ret = tracing_check_open_get_tr(file->tr);
4975 if (ret)
4976 return ret;
4977
4978 mutex_lock(&event_mutex);
4979
4980 /* Fail if the file is marked for removal */
4981 if (file->flags & EVENT_FILE_FL_FREED) {
4982 trace_array_put(file->tr);
4983 ret = -ENODEV;
4984 } else {
4985 event_file_get(file);
4986 }
4987
4988 mutex_unlock(&event_mutex);
4989 if (ret)
4990 return ret;
4991
4992 filp->private_data = inode->i_private;
4993
4994 return 0;
4995 }
4996
tracing_release_file_tr(struct inode * inode,struct file * filp)4997 int tracing_release_file_tr(struct inode *inode, struct file *filp)
4998 {
4999 struct trace_event_file *file = inode->i_private;
5000
5001 trace_array_put(file->tr);
5002 event_file_put(file);
5003
5004 return 0;
5005 }
5006
tracing_single_release_file_tr(struct inode * inode,struct file * filp)5007 int tracing_single_release_file_tr(struct inode *inode, struct file *filp)
5008 {
5009 tracing_release_file_tr(inode, filp);
5010 return single_release(inode, filp);
5011 }
5012
tracing_mark_open(struct inode * inode,struct file * filp)5013 static int tracing_mark_open(struct inode *inode, struct file *filp)
5014 {
5015 stream_open(inode, filp);
5016 return tracing_open_generic_tr(inode, filp);
5017 }
5018
tracing_release(struct inode * inode,struct file * file)5019 static int tracing_release(struct inode *inode, struct file *file)
5020 {
5021 struct trace_array *tr = inode->i_private;
5022 struct seq_file *m = file->private_data;
5023 struct trace_iterator *iter;
5024 int cpu;
5025
5026 if (!(file->f_mode & FMODE_READ)) {
5027 trace_array_put(tr);
5028 return 0;
5029 }
5030
5031 /* Writes do not use seq_file */
5032 iter = m->private;
5033 mutex_lock(&trace_types_lock);
5034
5035 for_each_tracing_cpu(cpu) {
5036 if (iter->buffer_iter[cpu])
5037 ring_buffer_read_finish(iter->buffer_iter[cpu]);
5038 }
5039
5040 if (iter->trace && iter->trace->close)
5041 iter->trace->close(iter);
5042
5043 if (!iter->snapshot && tr->stop_count)
5044 /* reenable tracing if it was previously enabled */
5045 tracing_start_tr(tr);
5046
5047 __trace_array_put(tr);
5048
5049 mutex_unlock(&trace_types_lock);
5050
5051 free_trace_iter_content(iter);
5052 seq_release_private(inode, file);
5053
5054 return 0;
5055 }
5056
tracing_release_generic_tr(struct inode * inode,struct file * file)5057 int tracing_release_generic_tr(struct inode *inode, struct file *file)
5058 {
5059 struct trace_array *tr = inode->i_private;
5060
5061 trace_array_put(tr);
5062 return 0;
5063 }
5064
tracing_single_release_tr(struct inode * inode,struct file * file)5065 static int tracing_single_release_tr(struct inode *inode, struct file *file)
5066 {
5067 struct trace_array *tr = inode->i_private;
5068
5069 trace_array_put(tr);
5070
5071 return single_release(inode, file);
5072 }
5073
tracing_open(struct inode * inode,struct file * file)5074 static int tracing_open(struct inode *inode, struct file *file)
5075 {
5076 struct trace_array *tr = inode->i_private;
5077 struct trace_iterator *iter;
5078 int ret;
5079
5080 ret = tracing_check_open_get_tr(tr);
5081 if (ret)
5082 return ret;
5083
5084 /* If this file was open for write, then erase contents */
5085 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC)) {
5086 int cpu = tracing_get_cpu(inode);
5087 struct array_buffer *trace_buf = &tr->array_buffer;
5088
5089 #ifdef CONFIG_TRACER_MAX_TRACE
5090 if (tr->current_trace->print_max)
5091 trace_buf = &tr->max_buffer;
5092 #endif
5093
5094 if (cpu == RING_BUFFER_ALL_CPUS)
5095 tracing_reset_online_cpus(trace_buf);
5096 else
5097 tracing_reset_cpu(trace_buf, cpu);
5098 }
5099
5100 if (file->f_mode & FMODE_READ) {
5101 iter = __tracing_open(inode, file, false);
5102 if (IS_ERR(iter))
5103 ret = PTR_ERR(iter);
5104 else if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
5105 iter->iter_flags |= TRACE_FILE_LAT_FMT;
5106 }
5107
5108 if (ret < 0)
5109 trace_array_put(tr);
5110
5111 return ret;
5112 }
5113
5114 /*
5115 * Some tracers are not suitable for instance buffers.
5116 * A tracer is always available for the global array (toplevel)
5117 * or if it explicitly states that it is.
5118 */
5119 static bool
trace_ok_for_array(struct tracer * t,struct trace_array * tr)5120 trace_ok_for_array(struct tracer *t, struct trace_array *tr)
5121 {
5122 return (tr->flags & TRACE_ARRAY_FL_GLOBAL) || t->allow_instances;
5123 }
5124
5125 /* Find the next tracer that this trace array may use */
5126 static struct tracer *
get_tracer_for_array(struct trace_array * tr,struct tracer * t)5127 get_tracer_for_array(struct trace_array *tr, struct tracer *t)
5128 {
5129 while (t && !trace_ok_for_array(t, tr))
5130 t = t->next;
5131
5132 return t;
5133 }
5134
5135 static void *
t_next(struct seq_file * m,void * v,loff_t * pos)5136 t_next(struct seq_file *m, void *v, loff_t *pos)
5137 {
5138 struct trace_array *tr = m->private;
5139 struct tracer *t = v;
5140
5141 (*pos)++;
5142
5143 if (t)
5144 t = get_tracer_for_array(tr, t->next);
5145
5146 return t;
5147 }
5148
t_start(struct seq_file * m,loff_t * pos)5149 static void *t_start(struct seq_file *m, loff_t *pos)
5150 {
5151 struct trace_array *tr = m->private;
5152 struct tracer *t;
5153 loff_t l = 0;
5154
5155 mutex_lock(&trace_types_lock);
5156
5157 t = get_tracer_for_array(tr, trace_types);
5158 for (; t && l < *pos; t = t_next(m, t, &l))
5159 ;
5160
5161 return t;
5162 }
5163
t_stop(struct seq_file * m,void * p)5164 static void t_stop(struct seq_file *m, void *p)
5165 {
5166 mutex_unlock(&trace_types_lock);
5167 }
5168
t_show(struct seq_file * m,void * v)5169 static int t_show(struct seq_file *m, void *v)
5170 {
5171 struct tracer *t = v;
5172
5173 if (!t)
5174 return 0;
5175
5176 seq_puts(m, t->name);
5177 if (t->next)
5178 seq_putc(m, ' ');
5179 else
5180 seq_putc(m, '\n');
5181
5182 return 0;
5183 }
5184
5185 static const struct seq_operations show_traces_seq_ops = {
5186 .start = t_start,
5187 .next = t_next,
5188 .stop = t_stop,
5189 .show = t_show,
5190 };
5191
show_traces_open(struct inode * inode,struct file * file)5192 static int show_traces_open(struct inode *inode, struct file *file)
5193 {
5194 struct trace_array *tr = inode->i_private;
5195 struct seq_file *m;
5196 int ret;
5197
5198 ret = tracing_check_open_get_tr(tr);
5199 if (ret)
5200 return ret;
5201
5202 ret = seq_open(file, &show_traces_seq_ops);
5203 if (ret) {
5204 trace_array_put(tr);
5205 return ret;
5206 }
5207
5208 m = file->private_data;
5209 m->private = tr;
5210
5211 return 0;
5212 }
5213
show_traces_release(struct inode * inode,struct file * file)5214 static int show_traces_release(struct inode *inode, struct file *file)
5215 {
5216 struct trace_array *tr = inode->i_private;
5217
5218 trace_array_put(tr);
5219 return seq_release(inode, file);
5220 }
5221
5222 static ssize_t
tracing_write_stub(struct file * filp,const char __user * ubuf,size_t count,loff_t * ppos)5223 tracing_write_stub(struct file *filp, const char __user *ubuf,
5224 size_t count, loff_t *ppos)
5225 {
5226 return count;
5227 }
5228
tracing_lseek(struct file * file,loff_t offset,int whence)5229 loff_t tracing_lseek(struct file *file, loff_t offset, int whence)
5230 {
5231 int ret;
5232
5233 if (file->f_mode & FMODE_READ)
5234 ret = seq_lseek(file, offset, whence);
5235 else
5236 file->f_pos = ret = 0;
5237
5238 return ret;
5239 }
5240
5241 static const struct file_operations tracing_fops = {
5242 .open = tracing_open,
5243 .read = seq_read,
5244 .read_iter = seq_read_iter,
5245 .splice_read = copy_splice_read,
5246 .write = tracing_write_stub,
5247 .llseek = tracing_lseek,
5248 .release = tracing_release,
5249 };
5250
5251 static const struct file_operations show_traces_fops = {
5252 .open = show_traces_open,
5253 .read = seq_read,
5254 .llseek = seq_lseek,
5255 .release = show_traces_release,
5256 };
5257
5258 static ssize_t
tracing_cpumask_read(struct file * filp,char __user * ubuf,size_t count,loff_t * ppos)5259 tracing_cpumask_read(struct file *filp, char __user *ubuf,
5260 size_t count, loff_t *ppos)
5261 {
5262 struct trace_array *tr = file_inode(filp)->i_private;
5263 char *mask_str;
5264 int len;
5265
5266 len = snprintf(NULL, 0, "%*pb\n",
5267 cpumask_pr_args(tr->tracing_cpumask)) + 1;
5268 mask_str = kmalloc(len, GFP_KERNEL);
5269 if (!mask_str)
5270 return -ENOMEM;
5271
5272 len = snprintf(mask_str, len, "%*pb\n",
5273 cpumask_pr_args(tr->tracing_cpumask));
5274 if (len >= count) {
5275 count = -EINVAL;
5276 goto out_err;
5277 }
5278 count = simple_read_from_buffer(ubuf, count, ppos, mask_str, len);
5279
5280 out_err:
5281 kfree(mask_str);
5282
5283 return count;
5284 }
5285
tracing_set_cpumask(struct trace_array * tr,cpumask_var_t tracing_cpumask_new)5286 int tracing_set_cpumask(struct trace_array *tr,
5287 cpumask_var_t tracing_cpumask_new)
5288 {
5289 int cpu;
5290
5291 if (!tr)
5292 return -EINVAL;
5293
5294 local_irq_disable();
5295 arch_spin_lock(&tr->max_lock);
5296 for_each_tracing_cpu(cpu) {
5297 /*
5298 * Increase/decrease the disabled counter if we are
5299 * about to flip a bit in the cpumask:
5300 */
5301 if (cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5302 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5303 atomic_inc(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5304 ring_buffer_record_disable_cpu(tr->array_buffer.buffer, cpu);
5305 #ifdef CONFIG_TRACER_MAX_TRACE
5306 ring_buffer_record_disable_cpu(tr->max_buffer.buffer, cpu);
5307 #endif
5308 }
5309 if (!cpumask_test_cpu(cpu, tr->tracing_cpumask) &&
5310 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
5311 atomic_dec(&per_cpu_ptr(tr->array_buffer.data, cpu)->disabled);
5312 ring_buffer_record_enable_cpu(tr->array_buffer.buffer, cpu);
5313 #ifdef CONFIG_TRACER_MAX_TRACE
5314 ring_buffer_record_enable_cpu(tr->max_buffer.buffer, cpu);
5315 #endif
5316 }
5317 }
5318 arch_spin_unlock(&tr->max_lock);
5319 local_irq_enable();
5320
5321 cpumask_copy(tr->tracing_cpumask, tracing_cpumask_new);
5322
5323 return 0;
5324 }
5325
5326 static ssize_t
tracing_cpumask_write(struct file * filp,const char __user * ubuf,size_t count,loff_t * ppos)5327 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
5328 size_t count, loff_t *ppos)
5329 {
5330 struct trace_array *tr = file_inode(filp)->i_private;
5331 cpumask_var_t tracing_cpumask_new;
5332 int err;
5333
5334 if (!zalloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
5335 return -ENOMEM;
5336
5337 err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
5338 if (err)
5339 goto err_free;
5340
5341 err = tracing_set_cpumask(tr, tracing_cpumask_new);
5342 if (err)
5343 goto err_free;
5344
5345 free_cpumask_var(tracing_cpumask_new);
5346
5347 return count;
5348
5349 err_free:
5350 free_cpumask_var(tracing_cpumask_new);
5351
5352 return err;
5353 }
5354
5355 static const struct file_operations tracing_cpumask_fops = {
5356 .open = tracing_open_generic_tr,
5357 .read = tracing_cpumask_read,
5358 .write = tracing_cpumask_write,
5359 .release = tracing_release_generic_tr,
5360 .llseek = generic_file_llseek,
5361 };
5362
tracing_trace_options_show(struct seq_file * m,void * v)5363 static int tracing_trace_options_show(struct seq_file *m, void *v)
5364 {
5365 struct tracer_opt *trace_opts;
5366 struct trace_array *tr = m->private;
5367 u32 tracer_flags;
5368 int i;
5369
5370 mutex_lock(&trace_types_lock);
5371 tracer_flags = tr->current_trace->flags->val;
5372 trace_opts = tr->current_trace->flags->opts;
5373
5374 for (i = 0; trace_options[i]; i++) {
5375 if (tr->trace_flags & (1 << i))
5376 seq_printf(m, "%s\n", trace_options[i]);
5377 else
5378 seq_printf(m, "no%s\n", trace_options[i]);
5379 }
5380
5381 for (i = 0; trace_opts[i].name; i++) {
5382 if (tracer_flags & trace_opts[i].bit)
5383 seq_printf(m, "%s\n", trace_opts[i].name);
5384 else
5385 seq_printf(m, "no%s\n", trace_opts[i].name);
5386 }
5387 mutex_unlock(&trace_types_lock);
5388
5389 return 0;
5390 }
5391
__set_tracer_option(struct trace_array * tr,struct tracer_flags * tracer_flags,struct tracer_opt * opts,int neg)5392 static int __set_tracer_option(struct trace_array *tr,
5393 struct tracer_flags *tracer_flags,
5394 struct tracer_opt *opts, int neg)
5395 {
5396 struct tracer *trace = tracer_flags->trace;
5397 int ret;
5398
5399 ret = trace->set_flag(tr, tracer_flags->val, opts->bit, !neg);
5400 if (ret)
5401 return ret;
5402
5403 if (neg)
5404 tracer_flags->val &= ~opts->bit;
5405 else
5406 tracer_flags->val |= opts->bit;
5407 return 0;
5408 }
5409
5410 /* Try to assign a tracer specific option */
set_tracer_option(struct trace_array * tr,char * cmp,int neg)5411 static int set_tracer_option(struct trace_array *tr, char *cmp, int neg)
5412 {
5413 struct tracer *trace = tr->current_trace;
5414 struct tracer_flags *tracer_flags = trace->flags;
5415 struct tracer_opt *opts = NULL;
5416 int i;
5417
5418 for (i = 0; tracer_flags->opts[i].name; i++) {
5419 opts = &tracer_flags->opts[i];
5420
5421 if (strcmp(cmp, opts->name) == 0)
5422 return __set_tracer_option(tr, trace->flags, opts, neg);
5423 }
5424
5425 return -EINVAL;
5426 }
5427
5428 /* Some tracers require overwrite to stay enabled */
trace_keep_overwrite(struct tracer * tracer,u32 mask,int set)5429 int trace_keep_overwrite(struct tracer *tracer, u32 mask, int set)
5430 {
5431 if (tracer->enabled && (mask & TRACE_ITER_OVERWRITE) && !set)
5432 return -1;
5433
5434 return 0;
5435 }
5436
set_tracer_flag(struct trace_array * tr,unsigned int mask,int enabled)5437 int set_tracer_flag(struct trace_array *tr, unsigned int mask, int enabled)
5438 {
5439 int *map;
5440
5441 if ((mask == TRACE_ITER_RECORD_TGID) ||
5442 (mask == TRACE_ITER_RECORD_CMD))
5443 lockdep_assert_held(&event_mutex);
5444
5445 /* do nothing if flag is already set */
5446 if (!!(tr->trace_flags & mask) == !!enabled)
5447 return 0;
5448
5449 /* Give the tracer a chance to approve the change */
5450 if (tr->current_trace->flag_changed)
5451 if (tr->current_trace->flag_changed(tr, mask, !!enabled))
5452 return -EINVAL;
5453
5454 if (enabled)
5455 tr->trace_flags |= mask;
5456 else
5457 tr->trace_flags &= ~mask;
5458
5459 if (mask == TRACE_ITER_RECORD_CMD)
5460 trace_event_enable_cmd_record(enabled);
5461
5462 if (mask == TRACE_ITER_RECORD_TGID) {
5463 if (!tgid_map) {
5464 tgid_map_max = pid_max;
5465 map = kvcalloc(tgid_map_max + 1, sizeof(*tgid_map),
5466 GFP_KERNEL);
5467
5468 /*
5469 * Pairs with smp_load_acquire() in
5470 * trace_find_tgid_ptr() to ensure that if it observes
5471 * the tgid_map we just allocated then it also observes
5472 * the corresponding tgid_map_max value.
5473 */
5474 smp_store_release(&tgid_map, map);
5475 }
5476 if (!tgid_map) {
5477 tr->trace_flags &= ~TRACE_ITER_RECORD_TGID;
5478 return -ENOMEM;
5479 }
5480
5481 trace_event_enable_tgid_record(enabled);
5482 }
5483
5484 if (mask == TRACE_ITER_EVENT_FORK)
5485 trace_event_follow_fork(tr, enabled);
5486
5487 if (mask == TRACE_ITER_FUNC_FORK)
5488 ftrace_pid_follow_fork(tr, enabled);
5489
5490 if (mask == TRACE_ITER_OVERWRITE) {
5491 ring_buffer_change_overwrite(tr->array_buffer.buffer, enabled);
5492 #ifdef CONFIG_TRACER_MAX_TRACE
5493 ring_buffer_change_overwrite(tr->max_buffer.buffer, enabled);
5494 #endif
5495 }
5496
5497 if (mask == TRACE_ITER_PRINTK) {
5498 trace_printk_start_stop_comm(enabled);
5499 trace_printk_control(enabled);
5500 }
5501
5502 return 0;
5503 }
5504
trace_set_options(struct trace_array * tr,char * option)5505 int trace_set_options(struct trace_array *tr, char *option)
5506 {
5507 char *cmp;
5508 int neg = 0;
5509 int ret;
5510 size_t orig_len = strlen(option);
5511 int len;
5512
5513 cmp = strstrip(option);
5514
5515 len = str_has_prefix(cmp, "no");
5516 if (len)
5517 neg = 1;
5518
5519 cmp += len;
5520
5521 mutex_lock(&event_mutex);
5522 mutex_lock(&trace_types_lock);
5523
5524 ret = match_string(trace_options, -1, cmp);
5525 /* If no option could be set, test the specific tracer options */
5526 if (ret < 0)
5527 ret = set_tracer_option(tr, cmp, neg);
5528 else
5529 ret = set_tracer_flag(tr, 1 << ret, !neg);
5530
5531 mutex_unlock(&trace_types_lock);
5532 mutex_unlock(&event_mutex);
5533
5534 /*
5535 * If the first trailing whitespace is replaced with '\0' by strstrip,
5536 * turn it back into a space.
5537 */
5538 if (orig_len > strlen(option))
5539 option[strlen(option)] = ' ';
5540
5541 return ret;
5542 }
5543
apply_trace_boot_options(void)5544 static void __init apply_trace_boot_options(void)
5545 {
5546 char *buf = trace_boot_options_buf;
5547 char *option;
5548
5549 while (true) {
5550 option = strsep(&buf, ",");
5551
5552 if (!option)
5553 break;
5554
5555 if (*option)
5556 trace_set_options(&global_trace, option);
5557
5558 /* Put back the comma to allow this to be called again */
5559 if (buf)
5560 *(buf - 1) = ',';
5561 }
5562 }
5563
5564 static ssize_t
tracing_trace_options_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)5565 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
5566 size_t cnt, loff_t *ppos)
5567 {
5568 struct seq_file *m = filp->private_data;
5569 struct trace_array *tr = m->private;
5570 char buf[64];
5571 int ret;
5572
5573 if (cnt >= sizeof(buf))
5574 return -EINVAL;
5575
5576 if (copy_from_user(buf, ubuf, cnt))
5577 return -EFAULT;
5578
5579 buf[cnt] = 0;
5580
5581 ret = trace_set_options(tr, buf);
5582 if (ret < 0)
5583 return ret;
5584
5585 *ppos += cnt;
5586
5587 return cnt;
5588 }
5589
tracing_trace_options_open(struct inode * inode,struct file * file)5590 static int tracing_trace_options_open(struct inode *inode, struct file *file)
5591 {
5592 struct trace_array *tr = inode->i_private;
5593 int ret;
5594
5595 ret = tracing_check_open_get_tr(tr);
5596 if (ret)
5597 return ret;
5598
5599 ret = single_open(file, tracing_trace_options_show, inode->i_private);
5600 if (ret < 0)
5601 trace_array_put(tr);
5602
5603 return ret;
5604 }
5605
5606 static const struct file_operations tracing_iter_fops = {
5607 .open = tracing_trace_options_open,
5608 .read = seq_read,
5609 .llseek = seq_lseek,
5610 .release = tracing_single_release_tr,
5611 .write = tracing_trace_options_write,
5612 };
5613
5614 static const char readme_msg[] =
5615 "tracing mini-HOWTO:\n\n"
5616 "# echo 0 > tracing_on : quick way to disable tracing\n"
5617 "# echo 1 > tracing_on : quick way to re-enable tracing\n\n"
5618 " Important files:\n"
5619 " trace\t\t\t- The static contents of the buffer\n"
5620 "\t\t\t To clear the buffer write into this file: echo > trace\n"
5621 " trace_pipe\t\t- A consuming read to see the contents of the buffer\n"
5622 " current_tracer\t- function and latency tracers\n"
5623 " available_tracers\t- list of configured tracers for current_tracer\n"
5624 " error_log\t- error log for failed commands (that support it)\n"
5625 " buffer_size_kb\t- view and modify size of per cpu buffer\n"
5626 " buffer_total_size_kb - view total size of all cpu buffers\n\n"
5627 " trace_clock\t\t- change the clock used to order events\n"
5628 " local: Per cpu clock but may not be synced across CPUs\n"
5629 " global: Synced across CPUs but slows tracing down.\n"
5630 " counter: Not a clock, but just an increment\n"
5631 " uptime: Jiffy counter from time of boot\n"
5632 " perf: Same clock that perf events use\n"
5633 #ifdef CONFIG_X86_64
5634 " x86-tsc: TSC cycle counter\n"
5635 #endif
5636 "\n timestamp_mode\t- view the mode used to timestamp events\n"
5637 " delta: Delta difference against a buffer-wide timestamp\n"
5638 " absolute: Absolute (standalone) timestamp\n"
5639 "\n trace_marker\t\t- Writes into this file writes into the kernel buffer\n"
5640 "\n trace_marker_raw\t\t- Writes into this file writes binary data into the kernel buffer\n"
5641 " tracing_cpumask\t- Limit which CPUs to trace\n"
5642 " instances\t\t- Make sub-buffers with: mkdir instances/foo\n"
5643 "\t\t\t Remove sub-buffer with rmdir\n"
5644 " trace_options\t\t- Set format or modify how tracing happens\n"
5645 "\t\t\t Disable an option by prefixing 'no' to the\n"
5646 "\t\t\t option name\n"
5647 " saved_cmdlines_size\t- echo command number in here to store comm-pid list\n"
5648 #ifdef CONFIG_DYNAMIC_FTRACE
5649 "\n available_filter_functions - list of functions that can be filtered on\n"
5650 " set_ftrace_filter\t- echo function name in here to only trace these\n"
5651 "\t\t\t functions\n"
5652 "\t accepts: func_full_name or glob-matching-pattern\n"
5653 "\t modules: Can select a group via module\n"
5654 "\t Format: :mod:<module-name>\n"
5655 "\t example: echo :mod:ext3 > set_ftrace_filter\n"
5656 "\t triggers: a command to perform when function is hit\n"
5657 "\t Format: <function>:<trigger>[:count]\n"
5658 "\t trigger: traceon, traceoff\n"
5659 "\t\t enable_event:<system>:<event>\n"
5660 "\t\t disable_event:<system>:<event>\n"
5661 #ifdef CONFIG_STACKTRACE
5662 "\t\t stacktrace\n"
5663 #endif
5664 #ifdef CONFIG_TRACER_SNAPSHOT
5665 "\t\t snapshot\n"
5666 #endif
5667 "\t\t dump\n"
5668 "\t\t cpudump\n"
5669 "\t example: echo do_fault:traceoff > set_ftrace_filter\n"
5670 "\t echo do_trap:traceoff:3 > set_ftrace_filter\n"
5671 "\t The first one will disable tracing every time do_fault is hit\n"
5672 "\t The second will disable tracing at most 3 times when do_trap is hit\n"
5673 "\t The first time do trap is hit and it disables tracing, the\n"
5674 "\t counter will decrement to 2. If tracing is already disabled,\n"
5675 "\t the counter will not decrement. It only decrements when the\n"
5676 "\t trigger did work\n"
5677 "\t To remove trigger without count:\n"
5678 "\t echo '!<function>:<trigger> > set_ftrace_filter\n"
5679 "\t To remove trigger with a count:\n"
5680 "\t echo '!<function>:<trigger>:0 > set_ftrace_filter\n"
5681 " set_ftrace_notrace\t- echo function name in here to never trace.\n"
5682 "\t accepts: func_full_name, *func_end, func_begin*, *func_middle*\n"
5683 "\t modules: Can select a group via module command :mod:\n"
5684 "\t Does not accept triggers\n"
5685 #endif /* CONFIG_DYNAMIC_FTRACE */
5686 #ifdef CONFIG_FUNCTION_TRACER
5687 " set_ftrace_pid\t- Write pid(s) to only function trace those pids\n"
5688 "\t\t (function)\n"
5689 " set_ftrace_notrace_pid\t- Write pid(s) to not function trace those pids\n"
5690 "\t\t (function)\n"
5691 #endif
5692 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5693 " set_graph_function\t- Trace the nested calls of a function (function_graph)\n"
5694 " set_graph_notrace\t- Do not trace the nested calls of a function (function_graph)\n"
5695 " max_graph_depth\t- Trace a limited depth of nested calls (0 is unlimited)\n"
5696 #endif
5697 #ifdef CONFIG_TRACER_SNAPSHOT
5698 "\n snapshot\t\t- Like 'trace' but shows the content of the static\n"
5699 "\t\t\t snapshot buffer. Read the contents for more\n"
5700 "\t\t\t information\n"
5701 #endif
5702 #ifdef CONFIG_STACK_TRACER
5703 " stack_trace\t\t- Shows the max stack trace when active\n"
5704 " stack_max_size\t- Shows current max stack size that was traced\n"
5705 "\t\t\t Write into this file to reset the max size (trigger a\n"
5706 "\t\t\t new trace)\n"
5707 #ifdef CONFIG_DYNAMIC_FTRACE
5708 " stack_trace_filter\t- Like set_ftrace_filter but limits what stack_trace\n"
5709 "\t\t\t traces\n"
5710 #endif
5711 #endif /* CONFIG_STACK_TRACER */
5712 #ifdef CONFIG_DYNAMIC_EVENTS
5713 " dynamic_events\t\t- Create/append/remove/show the generic dynamic events\n"
5714 "\t\t\t Write into this file to define/undefine new trace events.\n"
5715 #endif
5716 #ifdef CONFIG_KPROBE_EVENTS
5717 " kprobe_events\t\t- Create/append/remove/show the kernel dynamic events\n"
5718 "\t\t\t Write into this file to define/undefine new trace events.\n"
5719 #endif
5720 #ifdef CONFIG_UPROBE_EVENTS
5721 " uprobe_events\t\t- Create/append/remove/show the userspace dynamic events\n"
5722 "\t\t\t Write into this file to define/undefine new trace events.\n"
5723 #endif
5724 #if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS) || \
5725 defined(CONFIG_FPROBE_EVENTS)
5726 "\t accepts: event-definitions (one definition per line)\n"
5727 #if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
5728 "\t Format: p[:[<group>/][<event>]] <place> [<args>]\n"
5729 "\t r[maxactive][:[<group>/][<event>]] <place> [<args>]\n"
5730 #endif
5731 #ifdef CONFIG_FPROBE_EVENTS
5732 "\t f[:[<group>/][<event>]] <func-name>[%return] [<args>]\n"
5733 "\t t[:[<group>/][<event>]] <tracepoint> [<args>]\n"
5734 #endif
5735 #ifdef CONFIG_HIST_TRIGGERS
5736 "\t s:[synthetic/]<event> <field> [<field>]\n"
5737 #endif
5738 "\t e[:[<group>/][<event>]] <attached-group>.<attached-event> [<args>] [if <filter>]\n"
5739 "\t -:[<group>/][<event>]\n"
5740 #ifdef CONFIG_KPROBE_EVENTS
5741 "\t place: [<module>:]<symbol>[+<offset>]|<memaddr>\n"
5742 "place (kretprobe): [<module>:]<symbol>[+<offset>]%return|<memaddr>\n"
5743 #endif
5744 #ifdef CONFIG_UPROBE_EVENTS
5745 " place (uprobe): <path>:<offset>[%return][(ref_ctr_offset)]\n"
5746 #endif
5747 "\t args: <name>=fetcharg[:type]\n"
5748 "\t fetcharg: (%<register>|$<efield>), @<address>, @<symbol>[+|-<offset>],\n"
5749 #ifdef CONFIG_HAVE_FUNCTION_ARG_ACCESS_API
5750 #ifdef CONFIG_PROBE_EVENTS_BTF_ARGS
5751 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5752 "\t <argname>[->field[->field|.field...]],\n"
5753 #else
5754 "\t $stack<index>, $stack, $retval, $comm, $arg<N>,\n"
5755 #endif
5756 #else
5757 "\t $stack<index>, $stack, $retval, $comm,\n"
5758 #endif
5759 "\t +|-[u]<offset>(<fetcharg>), \\imm-value, \\\"imm-string\"\n"
5760 "\t type: s8/16/32/64, u8/16/32/64, x8/16/32/64, char, string, symbol,\n"
5761 "\t b<bit-width>@<bit-offset>/<container-size>, ustring,\n"
5762 "\t symstr, <type>\\[<array-size>\\]\n"
5763 #ifdef CONFIG_HIST_TRIGGERS
5764 "\t field: <stype> <name>;\n"
5765 "\t stype: u8/u16/u32/u64, s8/s16/s32/s64, pid_t,\n"
5766 "\t [unsigned] char/int/long\n"
5767 #endif
5768 "\t efield: For event probes ('e' types), the field is on of the fields\n"
5769 "\t of the <attached-group>/<attached-event>.\n"
5770 #endif
5771 " events/\t\t- Directory containing all trace event subsystems:\n"
5772 " enable\t\t- Write 0/1 to enable/disable tracing of all events\n"
5773 " events/<system>/\t- Directory containing all trace events for <system>:\n"
5774 " enable\t\t- Write 0/1 to enable/disable tracing of all <system>\n"
5775 "\t\t\t events\n"
5776 " filter\t\t- If set, only events passing filter are traced\n"
5777 " events/<system>/<event>/\t- Directory containing control files for\n"
5778 "\t\t\t <event>:\n"
5779 " enable\t\t- Write 0/1 to enable/disable tracing of <event>\n"
5780 " filter\t\t- If set, only events passing filter are traced\n"
5781 " trigger\t\t- If set, a command to perform when event is hit\n"
5782 "\t Format: <trigger>[:count][if <filter>]\n"
5783 "\t trigger: traceon, traceoff\n"
5784 "\t enable_event:<system>:<event>\n"
5785 "\t disable_event:<system>:<event>\n"
5786 #ifdef CONFIG_HIST_TRIGGERS
5787 "\t enable_hist:<system>:<event>\n"
5788 "\t disable_hist:<system>:<event>\n"
5789 #endif
5790 #ifdef CONFIG_STACKTRACE
5791 "\t\t stacktrace\n"
5792 #endif
5793 #ifdef CONFIG_TRACER_SNAPSHOT
5794 "\t\t snapshot\n"
5795 #endif
5796 #ifdef CONFIG_HIST_TRIGGERS
5797 "\t\t hist (see below)\n"
5798 #endif
5799 "\t example: echo traceoff > events/block/block_unplug/trigger\n"
5800 "\t echo traceoff:3 > events/block/block_unplug/trigger\n"
5801 "\t echo 'enable_event:kmem:kmalloc:3 if nr_rq > 1' > \\\n"
5802 "\t events/block/block_unplug/trigger\n"
5803 "\t The first disables tracing every time block_unplug is hit.\n"
5804 "\t The second disables tracing the first 3 times block_unplug is hit.\n"
5805 "\t The third enables the kmalloc event the first 3 times block_unplug\n"
5806 "\t is hit and has value of greater than 1 for the 'nr_rq' event field.\n"
5807 "\t Like function triggers, the counter is only decremented if it\n"
5808 "\t enabled or disabled tracing.\n"
5809 "\t To remove a trigger without a count:\n"
5810 "\t echo '!<trigger> > <system>/<event>/trigger\n"
5811 "\t To remove a trigger with a count:\n"
5812 "\t echo '!<trigger>:0 > <system>/<event>/trigger\n"
5813 "\t Filters can be ignored when removing a trigger.\n"
5814 #ifdef CONFIG_HIST_TRIGGERS
5815 " hist trigger\t- If set, event hits are aggregated into a hash table\n"
5816 "\t Format: hist:keys=<field1[,field2,...]>\n"
5817 "\t [:<var1>=<field|var_ref|numeric_literal>[,<var2>=...]]\n"
5818 "\t [:values=<field1[,field2,...]>]\n"
5819 "\t [:sort=<field1[,field2,...]>]\n"
5820 "\t [:size=#entries]\n"
5821 "\t [:pause][:continue][:clear]\n"
5822 "\t [:name=histname1]\n"
5823 "\t [:nohitcount]\n"
5824 "\t [:<handler>.<action>]\n"
5825 "\t [if <filter>]\n\n"
5826 "\t Note, special fields can be used as well:\n"
5827 "\t common_timestamp - to record current timestamp\n"
5828 "\t common_cpu - to record the CPU the event happened on\n"
5829 "\n"
5830 "\t A hist trigger variable can be:\n"
5831 "\t - a reference to a field e.g. x=current_timestamp,\n"
5832 "\t - a reference to another variable e.g. y=$x,\n"
5833 "\t - a numeric literal: e.g. ms_per_sec=1000,\n"
5834 "\t - an arithmetic expression: e.g. time_secs=current_timestamp/1000\n"
5835 "\n"
5836 "\t hist trigger arithmetic expressions support addition(+), subtraction(-),\n"
5837 "\t multiplication(*) and division(/) operators. An operand can be either a\n"
5838 "\t variable reference, field or numeric literal.\n"
5839 "\n"
5840 "\t When a matching event is hit, an entry is added to a hash\n"
5841 "\t table using the key(s) and value(s) named, and the value of a\n"
5842 "\t sum called 'hitcount' is incremented. Keys and values\n"
5843 "\t correspond to fields in the event's format description. Keys\n"
5844 "\t can be any field, or the special string 'common_stacktrace'.\n"
5845 "\t Compound keys consisting of up to two fields can be specified\n"
5846 "\t by the 'keys' keyword. Values must correspond to numeric\n"
5847 "\t fields. Sort keys consisting of up to two fields can be\n"
5848 "\t specified using the 'sort' keyword. The sort direction can\n"
5849 "\t be modified by appending '.descending' or '.ascending' to a\n"
5850 "\t sort field. The 'size' parameter can be used to specify more\n"
5851 "\t or fewer than the default 2048 entries for the hashtable size.\n"
5852 "\t If a hist trigger is given a name using the 'name' parameter,\n"
5853 "\t its histogram data will be shared with other triggers of the\n"
5854 "\t same name, and trigger hits will update this common data.\n\n"
5855 "\t Reading the 'hist' file for the event will dump the hash\n"
5856 "\t table in its entirety to stdout. If there are multiple hist\n"
5857 "\t triggers attached to an event, there will be a table for each\n"
5858 "\t trigger in the output. The table displayed for a named\n"
5859 "\t trigger will be the same as any other instance having the\n"
5860 "\t same name. The default format used to display a given field\n"
5861 "\t can be modified by appending any of the following modifiers\n"
5862 "\t to the field name, as applicable:\n\n"
5863 "\t .hex display a number as a hex value\n"
5864 "\t .sym display an address as a symbol\n"
5865 "\t .sym-offset display an address as a symbol and offset\n"
5866 "\t .execname display a common_pid as a program name\n"
5867 "\t .syscall display a syscall id as a syscall name\n"
5868 "\t .log2 display log2 value rather than raw number\n"
5869 "\t .buckets=size display values in groups of size rather than raw number\n"
5870 "\t .usecs display a common_timestamp in microseconds\n"
5871 "\t .percent display a number of percentage value\n"
5872 "\t .graph display a bar-graph of a value\n\n"
5873 "\t The 'pause' parameter can be used to pause an existing hist\n"
5874 "\t trigger or to start a hist trigger but not log any events\n"
5875 "\t until told to do so. 'continue' can be used to start or\n"
5876 "\t restart a paused hist trigger.\n\n"
5877 "\t The 'clear' parameter will clear the contents of a running\n"
5878 "\t hist trigger and leave its current paused/active state\n"
5879 "\t unchanged.\n\n"
5880 "\t The 'nohitcount' (or NOHC) parameter will suppress display of\n"
5881 "\t raw hitcount in the histogram.\n\n"
5882 "\t The enable_hist and disable_hist triggers can be used to\n"
5883 "\t have one event conditionally start and stop another event's\n"
5884 "\t already-attached hist trigger. The syntax is analogous to\n"
5885 "\t the enable_event and disable_event triggers.\n\n"
5886 "\t Hist trigger handlers and actions are executed whenever a\n"
5887 "\t a histogram entry is added or updated. They take the form:\n\n"
5888 "\t <handler>.<action>\n\n"
5889 "\t The available handlers are:\n\n"
5890 "\t onmatch(matching.event) - invoke on addition or update\n"
5891 "\t onmax(var) - invoke if var exceeds current max\n"
5892 "\t onchange(var) - invoke action if var changes\n\n"
5893 "\t The available actions are:\n\n"
5894 "\t trace(<synthetic_event>,param list) - generate synthetic event\n"
5895 "\t save(field,...) - save current event fields\n"
5896 #ifdef CONFIG_TRACER_SNAPSHOT
5897 "\t snapshot() - snapshot the trace buffer\n\n"
5898 #endif
5899 #ifdef CONFIG_SYNTH_EVENTS
5900 " events/synthetic_events\t- Create/append/remove/show synthetic events\n"
5901 "\t Write into this file to define/undefine new synthetic events.\n"
5902 "\t example: echo 'myevent u64 lat; char name[]; long[] stack' >> synthetic_events\n"
5903 #endif
5904 #endif
5905 ;
5906
5907 static ssize_t
tracing_readme_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)5908 tracing_readme_read(struct file *filp, char __user *ubuf,
5909 size_t cnt, loff_t *ppos)
5910 {
5911 return simple_read_from_buffer(ubuf, cnt, ppos,
5912 readme_msg, strlen(readme_msg));
5913 }
5914
5915 static const struct file_operations tracing_readme_fops = {
5916 .open = tracing_open_generic,
5917 .read = tracing_readme_read,
5918 .llseek = generic_file_llseek,
5919 };
5920
saved_tgids_next(struct seq_file * m,void * v,loff_t * pos)5921 static void *saved_tgids_next(struct seq_file *m, void *v, loff_t *pos)
5922 {
5923 int pid = ++(*pos);
5924
5925 return trace_find_tgid_ptr(pid);
5926 }
5927
saved_tgids_start(struct seq_file * m,loff_t * pos)5928 static void *saved_tgids_start(struct seq_file *m, loff_t *pos)
5929 {
5930 int pid = *pos;
5931
5932 return trace_find_tgid_ptr(pid);
5933 }
5934
saved_tgids_stop(struct seq_file * m,void * v)5935 static void saved_tgids_stop(struct seq_file *m, void *v)
5936 {
5937 }
5938
saved_tgids_show(struct seq_file * m,void * v)5939 static int saved_tgids_show(struct seq_file *m, void *v)
5940 {
5941 int *entry = (int *)v;
5942 int pid = entry - tgid_map;
5943 int tgid = *entry;
5944
5945 if (tgid == 0)
5946 return SEQ_SKIP;
5947
5948 seq_printf(m, "%d %d\n", pid, tgid);
5949 return 0;
5950 }
5951
5952 static const struct seq_operations tracing_saved_tgids_seq_ops = {
5953 .start = saved_tgids_start,
5954 .stop = saved_tgids_stop,
5955 .next = saved_tgids_next,
5956 .show = saved_tgids_show,
5957 };
5958
tracing_saved_tgids_open(struct inode * inode,struct file * filp)5959 static int tracing_saved_tgids_open(struct inode *inode, struct file *filp)
5960 {
5961 int ret;
5962
5963 ret = tracing_check_open_get_tr(NULL);
5964 if (ret)
5965 return ret;
5966
5967 return seq_open(filp, &tracing_saved_tgids_seq_ops);
5968 }
5969
5970
5971 static const struct file_operations tracing_saved_tgids_fops = {
5972 .open = tracing_saved_tgids_open,
5973 .read = seq_read,
5974 .llseek = seq_lseek,
5975 .release = seq_release,
5976 };
5977
saved_cmdlines_next(struct seq_file * m,void * v,loff_t * pos)5978 static void *saved_cmdlines_next(struct seq_file *m, void *v, loff_t *pos)
5979 {
5980 unsigned int *ptr = v;
5981
5982 if (*pos || m->count)
5983 ptr++;
5984
5985 (*pos)++;
5986
5987 for (; ptr < &savedcmd->map_cmdline_to_pid[savedcmd->cmdline_num];
5988 ptr++) {
5989 if (*ptr == -1 || *ptr == NO_CMDLINE_MAP)
5990 continue;
5991
5992 return ptr;
5993 }
5994
5995 return NULL;
5996 }
5997
saved_cmdlines_start(struct seq_file * m,loff_t * pos)5998 static void *saved_cmdlines_start(struct seq_file *m, loff_t *pos)
5999 {
6000 void *v;
6001 loff_t l = 0;
6002
6003 preempt_disable();
6004 arch_spin_lock(&trace_cmdline_lock);
6005
6006 v = &savedcmd->map_cmdline_to_pid[0];
6007 while (l <= *pos) {
6008 v = saved_cmdlines_next(m, v, &l);
6009 if (!v)
6010 return NULL;
6011 }
6012
6013 return v;
6014 }
6015
saved_cmdlines_stop(struct seq_file * m,void * v)6016 static void saved_cmdlines_stop(struct seq_file *m, void *v)
6017 {
6018 arch_spin_unlock(&trace_cmdline_lock);
6019 preempt_enable();
6020 }
6021
saved_cmdlines_show(struct seq_file * m,void * v)6022 static int saved_cmdlines_show(struct seq_file *m, void *v)
6023 {
6024 char buf[TASK_COMM_LEN];
6025 unsigned int *pid = v;
6026
6027 __trace_find_cmdline(*pid, buf);
6028 seq_printf(m, "%d %s\n", *pid, buf);
6029 return 0;
6030 }
6031
6032 static const struct seq_operations tracing_saved_cmdlines_seq_ops = {
6033 .start = saved_cmdlines_start,
6034 .next = saved_cmdlines_next,
6035 .stop = saved_cmdlines_stop,
6036 .show = saved_cmdlines_show,
6037 };
6038
tracing_saved_cmdlines_open(struct inode * inode,struct file * filp)6039 static int tracing_saved_cmdlines_open(struct inode *inode, struct file *filp)
6040 {
6041 int ret;
6042
6043 ret = tracing_check_open_get_tr(NULL);
6044 if (ret)
6045 return ret;
6046
6047 return seq_open(filp, &tracing_saved_cmdlines_seq_ops);
6048 }
6049
6050 static const struct file_operations tracing_saved_cmdlines_fops = {
6051 .open = tracing_saved_cmdlines_open,
6052 .read = seq_read,
6053 .llseek = seq_lseek,
6054 .release = seq_release,
6055 };
6056
6057 static ssize_t
tracing_saved_cmdlines_size_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)6058 tracing_saved_cmdlines_size_read(struct file *filp, char __user *ubuf,
6059 size_t cnt, loff_t *ppos)
6060 {
6061 char buf[64];
6062 int r;
6063
6064 preempt_disable();
6065 arch_spin_lock(&trace_cmdline_lock);
6066 r = scnprintf(buf, sizeof(buf), "%u\n", savedcmd->cmdline_num);
6067 arch_spin_unlock(&trace_cmdline_lock);
6068 preempt_enable();
6069
6070 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6071 }
6072
tracing_resize_saved_cmdlines(unsigned int val)6073 static int tracing_resize_saved_cmdlines(unsigned int val)
6074 {
6075 struct saved_cmdlines_buffer *s, *savedcmd_temp;
6076
6077 s = allocate_cmdlines_buffer(val);
6078 if (!s)
6079 return -ENOMEM;
6080
6081 preempt_disable();
6082 arch_spin_lock(&trace_cmdline_lock);
6083 savedcmd_temp = savedcmd;
6084 savedcmd = s;
6085 arch_spin_unlock(&trace_cmdline_lock);
6086 preempt_enable();
6087 free_saved_cmdlines_buffer(savedcmd_temp);
6088
6089 return 0;
6090 }
6091
6092 static ssize_t
tracing_saved_cmdlines_size_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)6093 tracing_saved_cmdlines_size_write(struct file *filp, const char __user *ubuf,
6094 size_t cnt, loff_t *ppos)
6095 {
6096 unsigned long val;
6097 int ret;
6098
6099 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6100 if (ret)
6101 return ret;
6102
6103 /* must have at least 1 entry or less than PID_MAX_DEFAULT */
6104 if (!val || val > PID_MAX_DEFAULT)
6105 return -EINVAL;
6106
6107 ret = tracing_resize_saved_cmdlines((unsigned int)val);
6108 if (ret < 0)
6109 return ret;
6110
6111 *ppos += cnt;
6112
6113 return cnt;
6114 }
6115
6116 static const struct file_operations tracing_saved_cmdlines_size_fops = {
6117 .open = tracing_open_generic,
6118 .read = tracing_saved_cmdlines_size_read,
6119 .write = tracing_saved_cmdlines_size_write,
6120 };
6121
6122 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
6123 static union trace_eval_map_item *
update_eval_map(union trace_eval_map_item * ptr)6124 update_eval_map(union trace_eval_map_item *ptr)
6125 {
6126 if (!ptr->map.eval_string) {
6127 if (ptr->tail.next) {
6128 ptr = ptr->tail.next;
6129 /* Set ptr to the next real item (skip head) */
6130 ptr++;
6131 } else
6132 return NULL;
6133 }
6134 return ptr;
6135 }
6136
eval_map_next(struct seq_file * m,void * v,loff_t * pos)6137 static void *eval_map_next(struct seq_file *m, void *v, loff_t *pos)
6138 {
6139 union trace_eval_map_item *ptr = v;
6140
6141 /*
6142 * Paranoid! If ptr points to end, we don't want to increment past it.
6143 * This really should never happen.
6144 */
6145 (*pos)++;
6146 ptr = update_eval_map(ptr);
6147 if (WARN_ON_ONCE(!ptr))
6148 return NULL;
6149
6150 ptr++;
6151 ptr = update_eval_map(ptr);
6152
6153 return ptr;
6154 }
6155
eval_map_start(struct seq_file * m,loff_t * pos)6156 static void *eval_map_start(struct seq_file *m, loff_t *pos)
6157 {
6158 union trace_eval_map_item *v;
6159 loff_t l = 0;
6160
6161 mutex_lock(&trace_eval_mutex);
6162
6163 v = trace_eval_maps;
6164 if (v)
6165 v++;
6166
6167 while (v && l < *pos) {
6168 v = eval_map_next(m, v, &l);
6169 }
6170
6171 return v;
6172 }
6173
eval_map_stop(struct seq_file * m,void * v)6174 static void eval_map_stop(struct seq_file *m, void *v)
6175 {
6176 mutex_unlock(&trace_eval_mutex);
6177 }
6178
eval_map_show(struct seq_file * m,void * v)6179 static int eval_map_show(struct seq_file *m, void *v)
6180 {
6181 union trace_eval_map_item *ptr = v;
6182
6183 seq_printf(m, "%s %ld (%s)\n",
6184 ptr->map.eval_string, ptr->map.eval_value,
6185 ptr->map.system);
6186
6187 return 0;
6188 }
6189
6190 static const struct seq_operations tracing_eval_map_seq_ops = {
6191 .start = eval_map_start,
6192 .next = eval_map_next,
6193 .stop = eval_map_stop,
6194 .show = eval_map_show,
6195 };
6196
tracing_eval_map_open(struct inode * inode,struct file * filp)6197 static int tracing_eval_map_open(struct inode *inode, struct file *filp)
6198 {
6199 int ret;
6200
6201 ret = tracing_check_open_get_tr(NULL);
6202 if (ret)
6203 return ret;
6204
6205 return seq_open(filp, &tracing_eval_map_seq_ops);
6206 }
6207
6208 static const struct file_operations tracing_eval_map_fops = {
6209 .open = tracing_eval_map_open,
6210 .read = seq_read,
6211 .llseek = seq_lseek,
6212 .release = seq_release,
6213 };
6214
6215 static inline union trace_eval_map_item *
trace_eval_jmp_to_tail(union trace_eval_map_item * ptr)6216 trace_eval_jmp_to_tail(union trace_eval_map_item *ptr)
6217 {
6218 /* Return tail of array given the head */
6219 return ptr + ptr->head.length + 1;
6220 }
6221
6222 static void
trace_insert_eval_map_file(struct module * mod,struct trace_eval_map ** start,int len)6223 trace_insert_eval_map_file(struct module *mod, struct trace_eval_map **start,
6224 int len)
6225 {
6226 struct trace_eval_map **stop;
6227 struct trace_eval_map **map;
6228 union trace_eval_map_item *map_array;
6229 union trace_eval_map_item *ptr;
6230
6231 stop = start + len;
6232
6233 /*
6234 * The trace_eval_maps contains the map plus a head and tail item,
6235 * where the head holds the module and length of array, and the
6236 * tail holds a pointer to the next list.
6237 */
6238 map_array = kmalloc_array(len + 2, sizeof(*map_array), GFP_KERNEL);
6239 if (!map_array) {
6240 pr_warn("Unable to allocate trace eval mapping\n");
6241 return;
6242 }
6243
6244 mutex_lock(&trace_eval_mutex);
6245
6246 if (!trace_eval_maps)
6247 trace_eval_maps = map_array;
6248 else {
6249 ptr = trace_eval_maps;
6250 for (;;) {
6251 ptr = trace_eval_jmp_to_tail(ptr);
6252 if (!ptr->tail.next)
6253 break;
6254 ptr = ptr->tail.next;
6255
6256 }
6257 ptr->tail.next = map_array;
6258 }
6259 map_array->head.mod = mod;
6260 map_array->head.length = len;
6261 map_array++;
6262
6263 for (map = start; (unsigned long)map < (unsigned long)stop; map++) {
6264 map_array->map = **map;
6265 map_array++;
6266 }
6267 memset(map_array, 0, sizeof(*map_array));
6268
6269 mutex_unlock(&trace_eval_mutex);
6270 }
6271
trace_create_eval_file(struct dentry * d_tracer)6272 static void trace_create_eval_file(struct dentry *d_tracer)
6273 {
6274 trace_create_file("eval_map", TRACE_MODE_READ, d_tracer,
6275 NULL, &tracing_eval_map_fops);
6276 }
6277
6278 #else /* CONFIG_TRACE_EVAL_MAP_FILE */
trace_create_eval_file(struct dentry * d_tracer)6279 static inline void trace_create_eval_file(struct dentry *d_tracer) { }
trace_insert_eval_map_file(struct module * mod,struct trace_eval_map ** start,int len)6280 static inline void trace_insert_eval_map_file(struct module *mod,
6281 struct trace_eval_map **start, int len) { }
6282 #endif /* !CONFIG_TRACE_EVAL_MAP_FILE */
6283
trace_insert_eval_map(struct module * mod,struct trace_eval_map ** start,int len)6284 static void trace_insert_eval_map(struct module *mod,
6285 struct trace_eval_map **start, int len)
6286 {
6287 struct trace_eval_map **map;
6288
6289 if (len <= 0)
6290 return;
6291
6292 map = start;
6293
6294 trace_event_eval_update(map, len);
6295
6296 trace_insert_eval_map_file(mod, start, len);
6297 }
6298
6299 static ssize_t
tracing_set_trace_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)6300 tracing_set_trace_read(struct file *filp, char __user *ubuf,
6301 size_t cnt, loff_t *ppos)
6302 {
6303 struct trace_array *tr = filp->private_data;
6304 char buf[MAX_TRACER_SIZE+2];
6305 int r;
6306
6307 mutex_lock(&trace_types_lock);
6308 r = sprintf(buf, "%s\n", tr->current_trace->name);
6309 mutex_unlock(&trace_types_lock);
6310
6311 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6312 }
6313
tracer_init(struct tracer * t,struct trace_array * tr)6314 int tracer_init(struct tracer *t, struct trace_array *tr)
6315 {
6316 tracing_reset_online_cpus(&tr->array_buffer);
6317 return t->init(tr);
6318 }
6319
set_buffer_entries(struct array_buffer * buf,unsigned long val)6320 static void set_buffer_entries(struct array_buffer *buf, unsigned long val)
6321 {
6322 int cpu;
6323
6324 for_each_tracing_cpu(cpu)
6325 per_cpu_ptr(buf->data, cpu)->entries = val;
6326 }
6327
update_buffer_entries(struct array_buffer * buf,int cpu)6328 static void update_buffer_entries(struct array_buffer *buf, int cpu)
6329 {
6330 if (cpu == RING_BUFFER_ALL_CPUS) {
6331 set_buffer_entries(buf, ring_buffer_size(buf->buffer, 0));
6332 } else {
6333 per_cpu_ptr(buf->data, cpu)->entries = ring_buffer_size(buf->buffer, cpu);
6334 }
6335 }
6336
6337 #ifdef CONFIG_TRACER_MAX_TRACE
6338 /* resize @tr's buffer to the size of @size_tr's entries */
resize_buffer_duplicate_size(struct array_buffer * trace_buf,struct array_buffer * size_buf,int cpu_id)6339 static int resize_buffer_duplicate_size(struct array_buffer *trace_buf,
6340 struct array_buffer *size_buf, int cpu_id)
6341 {
6342 int cpu, ret = 0;
6343
6344 if (cpu_id == RING_BUFFER_ALL_CPUS) {
6345 for_each_tracing_cpu(cpu) {
6346 ret = ring_buffer_resize(trace_buf->buffer,
6347 per_cpu_ptr(size_buf->data, cpu)->entries, cpu);
6348 if (ret < 0)
6349 break;
6350 per_cpu_ptr(trace_buf->data, cpu)->entries =
6351 per_cpu_ptr(size_buf->data, cpu)->entries;
6352 }
6353 } else {
6354 ret = ring_buffer_resize(trace_buf->buffer,
6355 per_cpu_ptr(size_buf->data, cpu_id)->entries, cpu_id);
6356 if (ret == 0)
6357 per_cpu_ptr(trace_buf->data, cpu_id)->entries =
6358 per_cpu_ptr(size_buf->data, cpu_id)->entries;
6359 }
6360
6361 return ret;
6362 }
6363 #endif /* CONFIG_TRACER_MAX_TRACE */
6364
__tracing_resize_ring_buffer(struct trace_array * tr,unsigned long size,int cpu)6365 static int __tracing_resize_ring_buffer(struct trace_array *tr,
6366 unsigned long size, int cpu)
6367 {
6368 int ret;
6369
6370 /*
6371 * If kernel or user changes the size of the ring buffer
6372 * we use the size that was given, and we can forget about
6373 * expanding it later.
6374 */
6375 trace_set_ring_buffer_expanded(tr);
6376
6377 /* May be called before buffers are initialized */
6378 if (!tr->array_buffer.buffer)
6379 return 0;
6380
6381 /* Do not allow tracing while resizing ring buffer */
6382 tracing_stop_tr(tr);
6383
6384 ret = ring_buffer_resize(tr->array_buffer.buffer, size, cpu);
6385 if (ret < 0)
6386 goto out_start;
6387
6388 #ifdef CONFIG_TRACER_MAX_TRACE
6389 if (!tr->allocated_snapshot)
6390 goto out;
6391
6392 ret = ring_buffer_resize(tr->max_buffer.buffer, size, cpu);
6393 if (ret < 0) {
6394 int r = resize_buffer_duplicate_size(&tr->array_buffer,
6395 &tr->array_buffer, cpu);
6396 if (r < 0) {
6397 /*
6398 * AARGH! We are left with different
6399 * size max buffer!!!!
6400 * The max buffer is our "snapshot" buffer.
6401 * When a tracer needs a snapshot (one of the
6402 * latency tracers), it swaps the max buffer
6403 * with the saved snap shot. We succeeded to
6404 * update the size of the main buffer, but failed to
6405 * update the size of the max buffer. But when we tried
6406 * to reset the main buffer to the original size, we
6407 * failed there too. This is very unlikely to
6408 * happen, but if it does, warn and kill all
6409 * tracing.
6410 */
6411 WARN_ON(1);
6412 tracing_disabled = 1;
6413 }
6414 goto out_start;
6415 }
6416
6417 update_buffer_entries(&tr->max_buffer, cpu);
6418
6419 out:
6420 #endif /* CONFIG_TRACER_MAX_TRACE */
6421
6422 update_buffer_entries(&tr->array_buffer, cpu);
6423 out_start:
6424 tracing_start_tr(tr);
6425 return ret;
6426 }
6427
tracing_resize_ring_buffer(struct trace_array * tr,unsigned long size,int cpu_id)6428 ssize_t tracing_resize_ring_buffer(struct trace_array *tr,
6429 unsigned long size, int cpu_id)
6430 {
6431 int ret;
6432
6433 mutex_lock(&trace_types_lock);
6434
6435 if (cpu_id != RING_BUFFER_ALL_CPUS) {
6436 /* make sure, this cpu is enabled in the mask */
6437 if (!cpumask_test_cpu(cpu_id, tracing_buffer_mask)) {
6438 ret = -EINVAL;
6439 goto out;
6440 }
6441 }
6442
6443 ret = __tracing_resize_ring_buffer(tr, size, cpu_id);
6444 if (ret < 0)
6445 ret = -ENOMEM;
6446
6447 out:
6448 mutex_unlock(&trace_types_lock);
6449
6450 return ret;
6451 }
6452
6453
6454 /**
6455 * tracing_update_buffers - used by tracing facility to expand ring buffers
6456 * @tr: The tracing instance
6457 *
6458 * To save on memory when the tracing is never used on a system with it
6459 * configured in. The ring buffers are set to a minimum size. But once
6460 * a user starts to use the tracing facility, then they need to grow
6461 * to their default size.
6462 *
6463 * This function is to be called when a tracer is about to be used.
6464 */
tracing_update_buffers(struct trace_array * tr)6465 int tracing_update_buffers(struct trace_array *tr)
6466 {
6467 int ret = 0;
6468
6469 mutex_lock(&trace_types_lock);
6470 if (!tr->ring_buffer_expanded)
6471 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6472 RING_BUFFER_ALL_CPUS);
6473 mutex_unlock(&trace_types_lock);
6474
6475 return ret;
6476 }
6477
6478 struct trace_option_dentry;
6479
6480 static void
6481 create_trace_option_files(struct trace_array *tr, struct tracer *tracer);
6482
6483 /*
6484 * Used to clear out the tracer before deletion of an instance.
6485 * Must have trace_types_lock held.
6486 */
tracing_set_nop(struct trace_array * tr)6487 static void tracing_set_nop(struct trace_array *tr)
6488 {
6489 if (tr->current_trace == &nop_trace)
6490 return;
6491
6492 tr->current_trace->enabled--;
6493
6494 if (tr->current_trace->reset)
6495 tr->current_trace->reset(tr);
6496
6497 tr->current_trace = &nop_trace;
6498 }
6499
6500 static bool tracer_options_updated;
6501
add_tracer_options(struct trace_array * tr,struct tracer * t)6502 static void add_tracer_options(struct trace_array *tr, struct tracer *t)
6503 {
6504 /* Only enable if the directory has been created already. */
6505 if (!tr->dir)
6506 return;
6507
6508 /* Only create trace option files after update_tracer_options finish */
6509 if (!tracer_options_updated)
6510 return;
6511
6512 create_trace_option_files(tr, t);
6513 }
6514
tracing_set_tracer(struct trace_array * tr,const char * buf)6515 int tracing_set_tracer(struct trace_array *tr, const char *buf)
6516 {
6517 struct tracer *t;
6518 #ifdef CONFIG_TRACER_MAX_TRACE
6519 bool had_max_tr;
6520 #endif
6521 int ret = 0;
6522
6523 mutex_lock(&trace_types_lock);
6524
6525 if (!tr->ring_buffer_expanded) {
6526 ret = __tracing_resize_ring_buffer(tr, trace_buf_size,
6527 RING_BUFFER_ALL_CPUS);
6528 if (ret < 0)
6529 goto out;
6530 ret = 0;
6531 }
6532
6533 for (t = trace_types; t; t = t->next) {
6534 if (strcmp(t->name, buf) == 0)
6535 break;
6536 }
6537 if (!t) {
6538 ret = -EINVAL;
6539 goto out;
6540 }
6541 if (t == tr->current_trace)
6542 goto out;
6543
6544 #ifdef CONFIG_TRACER_SNAPSHOT
6545 if (t->use_max_tr) {
6546 local_irq_disable();
6547 arch_spin_lock(&tr->max_lock);
6548 if (tr->cond_snapshot)
6549 ret = -EBUSY;
6550 arch_spin_unlock(&tr->max_lock);
6551 local_irq_enable();
6552 if (ret)
6553 goto out;
6554 }
6555 #endif
6556 /* Some tracers won't work on kernel command line */
6557 if (system_state < SYSTEM_RUNNING && t->noboot) {
6558 pr_warn("Tracer '%s' is not allowed on command line, ignored\n",
6559 t->name);
6560 goto out;
6561 }
6562
6563 /* Some tracers are only allowed for the top level buffer */
6564 if (!trace_ok_for_array(t, tr)) {
6565 ret = -EINVAL;
6566 goto out;
6567 }
6568
6569 /* If trace pipe files are being read, we can't change the tracer */
6570 if (tr->trace_ref) {
6571 ret = -EBUSY;
6572 goto out;
6573 }
6574
6575 trace_branch_disable();
6576
6577 tr->current_trace->enabled--;
6578
6579 if (tr->current_trace->reset)
6580 tr->current_trace->reset(tr);
6581
6582 #ifdef CONFIG_TRACER_MAX_TRACE
6583 had_max_tr = tr->current_trace->use_max_tr;
6584
6585 /* Current trace needs to be nop_trace before synchronize_rcu */
6586 tr->current_trace = &nop_trace;
6587
6588 if (had_max_tr && !t->use_max_tr) {
6589 /*
6590 * We need to make sure that the update_max_tr sees that
6591 * current_trace changed to nop_trace to keep it from
6592 * swapping the buffers after we resize it.
6593 * The update_max_tr is called from interrupts disabled
6594 * so a synchronized_sched() is sufficient.
6595 */
6596 synchronize_rcu();
6597 free_snapshot(tr);
6598 }
6599
6600 if (t->use_max_tr && !tr->allocated_snapshot) {
6601 ret = tracing_alloc_snapshot_instance(tr);
6602 if (ret < 0)
6603 goto out;
6604 }
6605 #else
6606 tr->current_trace = &nop_trace;
6607 #endif
6608
6609 if (t->init) {
6610 ret = tracer_init(t, tr);
6611 if (ret)
6612 goto out;
6613 }
6614
6615 tr->current_trace = t;
6616 tr->current_trace->enabled++;
6617 trace_branch_enable(tr);
6618 out:
6619 mutex_unlock(&trace_types_lock);
6620
6621 return ret;
6622 }
6623
6624 static ssize_t
tracing_set_trace_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)6625 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
6626 size_t cnt, loff_t *ppos)
6627 {
6628 struct trace_array *tr = filp->private_data;
6629 char buf[MAX_TRACER_SIZE+1];
6630 char *name;
6631 size_t ret;
6632 int err;
6633
6634 ret = cnt;
6635
6636 if (cnt > MAX_TRACER_SIZE)
6637 cnt = MAX_TRACER_SIZE;
6638
6639 if (copy_from_user(buf, ubuf, cnt))
6640 return -EFAULT;
6641
6642 buf[cnt] = 0;
6643
6644 name = strim(buf);
6645
6646 err = tracing_set_tracer(tr, name);
6647 if (err)
6648 return err;
6649
6650 *ppos += ret;
6651
6652 return ret;
6653 }
6654
6655 static ssize_t
tracing_nsecs_read(unsigned long * ptr,char __user * ubuf,size_t cnt,loff_t * ppos)6656 tracing_nsecs_read(unsigned long *ptr, char __user *ubuf,
6657 size_t cnt, loff_t *ppos)
6658 {
6659 char buf[64];
6660 int r;
6661
6662 r = snprintf(buf, sizeof(buf), "%ld\n",
6663 *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
6664 if (r > sizeof(buf))
6665 r = sizeof(buf);
6666 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
6667 }
6668
6669 static ssize_t
tracing_nsecs_write(unsigned long * ptr,const char __user * ubuf,size_t cnt,loff_t * ppos)6670 tracing_nsecs_write(unsigned long *ptr, const char __user *ubuf,
6671 size_t cnt, loff_t *ppos)
6672 {
6673 unsigned long val;
6674 int ret;
6675
6676 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
6677 if (ret)
6678 return ret;
6679
6680 *ptr = val * 1000;
6681
6682 return cnt;
6683 }
6684
6685 static ssize_t
tracing_thresh_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)6686 tracing_thresh_read(struct file *filp, char __user *ubuf,
6687 size_t cnt, loff_t *ppos)
6688 {
6689 return tracing_nsecs_read(&tracing_thresh, ubuf, cnt, ppos);
6690 }
6691
6692 static ssize_t
tracing_thresh_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)6693 tracing_thresh_write(struct file *filp, const char __user *ubuf,
6694 size_t cnt, loff_t *ppos)
6695 {
6696 struct trace_array *tr = filp->private_data;
6697 int ret;
6698
6699 mutex_lock(&trace_types_lock);
6700 ret = tracing_nsecs_write(&tracing_thresh, ubuf, cnt, ppos);
6701 if (ret < 0)
6702 goto out;
6703
6704 if (tr->current_trace->update_thresh) {
6705 ret = tr->current_trace->update_thresh(tr);
6706 if (ret < 0)
6707 goto out;
6708 }
6709
6710 ret = cnt;
6711 out:
6712 mutex_unlock(&trace_types_lock);
6713
6714 return ret;
6715 }
6716
6717 #ifdef CONFIG_TRACER_MAX_TRACE
6718
6719 static ssize_t
tracing_max_lat_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)6720 tracing_max_lat_read(struct file *filp, char __user *ubuf,
6721 size_t cnt, loff_t *ppos)
6722 {
6723 struct trace_array *tr = filp->private_data;
6724
6725 return tracing_nsecs_read(&tr->max_latency, ubuf, cnt, ppos);
6726 }
6727
6728 static ssize_t
tracing_max_lat_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)6729 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
6730 size_t cnt, loff_t *ppos)
6731 {
6732 struct trace_array *tr = filp->private_data;
6733
6734 return tracing_nsecs_write(&tr->max_latency, ubuf, cnt, ppos);
6735 }
6736
6737 #endif
6738
open_pipe_on_cpu(struct trace_array * tr,int cpu)6739 static int open_pipe_on_cpu(struct trace_array *tr, int cpu)
6740 {
6741 if (cpu == RING_BUFFER_ALL_CPUS) {
6742 if (cpumask_empty(tr->pipe_cpumask)) {
6743 cpumask_setall(tr->pipe_cpumask);
6744 return 0;
6745 }
6746 } else if (!cpumask_test_cpu(cpu, tr->pipe_cpumask)) {
6747 cpumask_set_cpu(cpu, tr->pipe_cpumask);
6748 return 0;
6749 }
6750 return -EBUSY;
6751 }
6752
close_pipe_on_cpu(struct trace_array * tr,int cpu)6753 static void close_pipe_on_cpu(struct trace_array *tr, int cpu)
6754 {
6755 if (cpu == RING_BUFFER_ALL_CPUS) {
6756 WARN_ON(!cpumask_full(tr->pipe_cpumask));
6757 cpumask_clear(tr->pipe_cpumask);
6758 } else {
6759 WARN_ON(!cpumask_test_cpu(cpu, tr->pipe_cpumask));
6760 cpumask_clear_cpu(cpu, tr->pipe_cpumask);
6761 }
6762 }
6763
tracing_open_pipe(struct inode * inode,struct file * filp)6764 static int tracing_open_pipe(struct inode *inode, struct file *filp)
6765 {
6766 struct trace_array *tr = inode->i_private;
6767 struct trace_iterator *iter;
6768 int cpu;
6769 int ret;
6770
6771 ret = tracing_check_open_get_tr(tr);
6772 if (ret)
6773 return ret;
6774
6775 mutex_lock(&trace_types_lock);
6776 cpu = tracing_get_cpu(inode);
6777 ret = open_pipe_on_cpu(tr, cpu);
6778 if (ret)
6779 goto fail_pipe_on_cpu;
6780
6781 /* create a buffer to store the information to pass to userspace */
6782 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
6783 if (!iter) {
6784 ret = -ENOMEM;
6785 goto fail_alloc_iter;
6786 }
6787
6788 trace_seq_init(&iter->seq);
6789 iter->trace = tr->current_trace;
6790
6791 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
6792 ret = -ENOMEM;
6793 goto fail;
6794 }
6795
6796 /* trace pipe does not show start of buffer */
6797 cpumask_setall(iter->started);
6798
6799 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
6800 iter->iter_flags |= TRACE_FILE_LAT_FMT;
6801
6802 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
6803 if (trace_clocks[tr->clock_id].in_ns)
6804 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
6805
6806 iter->tr = tr;
6807 iter->array_buffer = &tr->array_buffer;
6808 iter->cpu_file = cpu;
6809 mutex_init(&iter->mutex);
6810 filp->private_data = iter;
6811
6812 if (iter->trace->pipe_open)
6813 iter->trace->pipe_open(iter);
6814
6815 nonseekable_open(inode, filp);
6816
6817 tr->trace_ref++;
6818
6819 mutex_unlock(&trace_types_lock);
6820 return ret;
6821
6822 fail:
6823 kfree(iter);
6824 fail_alloc_iter:
6825 close_pipe_on_cpu(tr, cpu);
6826 fail_pipe_on_cpu:
6827 __trace_array_put(tr);
6828 mutex_unlock(&trace_types_lock);
6829 return ret;
6830 }
6831
tracing_release_pipe(struct inode * inode,struct file * file)6832 static int tracing_release_pipe(struct inode *inode, struct file *file)
6833 {
6834 struct trace_iterator *iter = file->private_data;
6835 struct trace_array *tr = inode->i_private;
6836
6837 mutex_lock(&trace_types_lock);
6838
6839 tr->trace_ref--;
6840
6841 if (iter->trace->pipe_close)
6842 iter->trace->pipe_close(iter);
6843 close_pipe_on_cpu(tr, iter->cpu_file);
6844 mutex_unlock(&trace_types_lock);
6845
6846 free_trace_iter_content(iter);
6847 kfree(iter);
6848
6849 trace_array_put(tr);
6850
6851 return 0;
6852 }
6853
6854 static __poll_t
trace_poll(struct trace_iterator * iter,struct file * filp,poll_table * poll_table)6855 trace_poll(struct trace_iterator *iter, struct file *filp, poll_table *poll_table)
6856 {
6857 struct trace_array *tr = iter->tr;
6858
6859 /* Iterators are static, they should be filled or empty */
6860 if (trace_buffer_iter(iter, iter->cpu_file))
6861 return EPOLLIN | EPOLLRDNORM;
6862
6863 if (tr->trace_flags & TRACE_ITER_BLOCK)
6864 /*
6865 * Always select as readable when in blocking mode
6866 */
6867 return EPOLLIN | EPOLLRDNORM;
6868 else
6869 return ring_buffer_poll_wait(iter->array_buffer->buffer, iter->cpu_file,
6870 filp, poll_table, iter->tr->buffer_percent);
6871 }
6872
6873 static __poll_t
tracing_poll_pipe(struct file * filp,poll_table * poll_table)6874 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
6875 {
6876 struct trace_iterator *iter = filp->private_data;
6877
6878 return trace_poll(iter, filp, poll_table);
6879 }
6880
6881 /* Must be called with iter->mutex held. */
tracing_wait_pipe(struct file * filp)6882 static int tracing_wait_pipe(struct file *filp)
6883 {
6884 struct trace_iterator *iter = filp->private_data;
6885 int ret;
6886
6887 while (trace_empty(iter)) {
6888
6889 if ((filp->f_flags & O_NONBLOCK)) {
6890 return -EAGAIN;
6891 }
6892
6893 /*
6894 * We block until we read something and tracing is disabled.
6895 * We still block if tracing is disabled, but we have never
6896 * read anything. This allows a user to cat this file, and
6897 * then enable tracing. But after we have read something,
6898 * we give an EOF when tracing is again disabled.
6899 *
6900 * iter->pos will be 0 if we haven't read anything.
6901 */
6902 if (!tracer_tracing_is_on(iter->tr) && iter->pos)
6903 break;
6904
6905 mutex_unlock(&iter->mutex);
6906
6907 ret = wait_on_pipe(iter, 0);
6908
6909 mutex_lock(&iter->mutex);
6910
6911 if (ret)
6912 return ret;
6913 }
6914
6915 return 1;
6916 }
6917
6918 /*
6919 * Consumer reader.
6920 */
6921 static ssize_t
tracing_read_pipe(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)6922 tracing_read_pipe(struct file *filp, char __user *ubuf,
6923 size_t cnt, loff_t *ppos)
6924 {
6925 struct trace_iterator *iter = filp->private_data;
6926 ssize_t sret;
6927
6928 /*
6929 * Avoid more than one consumer on a single file descriptor
6930 * This is just a matter of traces coherency, the ring buffer itself
6931 * is protected.
6932 */
6933 mutex_lock(&iter->mutex);
6934
6935 /* return any leftover data */
6936 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
6937 if (sret != -EBUSY)
6938 goto out;
6939
6940 trace_seq_init(&iter->seq);
6941
6942 if (iter->trace->read) {
6943 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
6944 if (sret)
6945 goto out;
6946 }
6947
6948 waitagain:
6949 sret = tracing_wait_pipe(filp);
6950 if (sret <= 0)
6951 goto out;
6952
6953 /* stop when tracing is finished */
6954 if (trace_empty(iter)) {
6955 sret = 0;
6956 goto out;
6957 }
6958
6959 if (cnt >= TRACE_SEQ_BUFFER_SIZE)
6960 cnt = TRACE_SEQ_BUFFER_SIZE - 1;
6961
6962 /* reset all but tr, trace, and overruns */
6963 trace_iterator_reset(iter);
6964 cpumask_clear(iter->started);
6965 trace_seq_init(&iter->seq);
6966
6967 trace_event_read_lock();
6968 trace_access_lock(iter->cpu_file);
6969 while (trace_find_next_entry_inc(iter) != NULL) {
6970 enum print_line_t ret;
6971 int save_len = iter->seq.seq.len;
6972
6973 ret = print_trace_line(iter);
6974 if (ret == TRACE_TYPE_PARTIAL_LINE) {
6975 /*
6976 * If one print_trace_line() fills entire trace_seq in one shot,
6977 * trace_seq_to_user() will returns -EBUSY because save_len == 0,
6978 * In this case, we need to consume it, otherwise, loop will peek
6979 * this event next time, resulting in an infinite loop.
6980 */
6981 if (save_len == 0) {
6982 iter->seq.full = 0;
6983 trace_seq_puts(&iter->seq, "[LINE TOO BIG]\n");
6984 trace_consume(iter);
6985 break;
6986 }
6987
6988 /* In other cases, don't print partial lines */
6989 iter->seq.seq.len = save_len;
6990 break;
6991 }
6992 if (ret != TRACE_TYPE_NO_CONSUME)
6993 trace_consume(iter);
6994
6995 if (trace_seq_used(&iter->seq) >= cnt)
6996 break;
6997
6998 /*
6999 * Setting the full flag means we reached the trace_seq buffer
7000 * size and we should leave by partial output condition above.
7001 * One of the trace_seq_* functions is not used properly.
7002 */
7003 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
7004 iter->ent->type);
7005 }
7006 trace_access_unlock(iter->cpu_file);
7007 trace_event_read_unlock();
7008
7009 /* Now copy what we have to the user */
7010 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
7011 if (iter->seq.readpos >= trace_seq_used(&iter->seq))
7012 trace_seq_init(&iter->seq);
7013
7014 /*
7015 * If there was nothing to send to user, in spite of consuming trace
7016 * entries, go back to wait for more entries.
7017 */
7018 if (sret == -EBUSY)
7019 goto waitagain;
7020
7021 out:
7022 mutex_unlock(&iter->mutex);
7023
7024 return sret;
7025 }
7026
tracing_spd_release_pipe(struct splice_pipe_desc * spd,unsigned int idx)7027 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
7028 unsigned int idx)
7029 {
7030 __free_page(spd->pages[idx]);
7031 }
7032
7033 static size_t
tracing_fill_pipe_page(size_t rem,struct trace_iterator * iter)7034 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
7035 {
7036 size_t count;
7037 int save_len;
7038 int ret;
7039
7040 /* Seq buffer is page-sized, exactly what we need. */
7041 for (;;) {
7042 save_len = iter->seq.seq.len;
7043 ret = print_trace_line(iter);
7044
7045 if (trace_seq_has_overflowed(&iter->seq)) {
7046 iter->seq.seq.len = save_len;
7047 break;
7048 }
7049
7050 /*
7051 * This should not be hit, because it should only
7052 * be set if the iter->seq overflowed. But check it
7053 * anyway to be safe.
7054 */
7055 if (ret == TRACE_TYPE_PARTIAL_LINE) {
7056 iter->seq.seq.len = save_len;
7057 break;
7058 }
7059
7060 count = trace_seq_used(&iter->seq) - save_len;
7061 if (rem < count) {
7062 rem = 0;
7063 iter->seq.seq.len = save_len;
7064 break;
7065 }
7066
7067 if (ret != TRACE_TYPE_NO_CONSUME)
7068 trace_consume(iter);
7069 rem -= count;
7070 if (!trace_find_next_entry_inc(iter)) {
7071 rem = 0;
7072 iter->ent = NULL;
7073 break;
7074 }
7075 }
7076
7077 return rem;
7078 }
7079
tracing_splice_read_pipe(struct file * filp,loff_t * ppos,struct pipe_inode_info * pipe,size_t len,unsigned int flags)7080 static ssize_t tracing_splice_read_pipe(struct file *filp,
7081 loff_t *ppos,
7082 struct pipe_inode_info *pipe,
7083 size_t len,
7084 unsigned int flags)
7085 {
7086 struct page *pages_def[PIPE_DEF_BUFFERS];
7087 struct partial_page partial_def[PIPE_DEF_BUFFERS];
7088 struct trace_iterator *iter = filp->private_data;
7089 struct splice_pipe_desc spd = {
7090 .pages = pages_def,
7091 .partial = partial_def,
7092 .nr_pages = 0, /* This gets updated below. */
7093 .nr_pages_max = PIPE_DEF_BUFFERS,
7094 .ops = &default_pipe_buf_ops,
7095 .spd_release = tracing_spd_release_pipe,
7096 };
7097 ssize_t ret;
7098 size_t rem;
7099 unsigned int i;
7100
7101 if (splice_grow_spd(pipe, &spd))
7102 return -ENOMEM;
7103
7104 mutex_lock(&iter->mutex);
7105
7106 if (iter->trace->splice_read) {
7107 ret = iter->trace->splice_read(iter, filp,
7108 ppos, pipe, len, flags);
7109 if (ret)
7110 goto out_err;
7111 }
7112
7113 ret = tracing_wait_pipe(filp);
7114 if (ret <= 0)
7115 goto out_err;
7116
7117 if (!iter->ent && !trace_find_next_entry_inc(iter)) {
7118 ret = -EFAULT;
7119 goto out_err;
7120 }
7121
7122 trace_event_read_lock();
7123 trace_access_lock(iter->cpu_file);
7124
7125 /* Fill as many pages as possible. */
7126 for (i = 0, rem = len; i < spd.nr_pages_max && rem; i++) {
7127 spd.pages[i] = alloc_page(GFP_KERNEL);
7128 if (!spd.pages[i])
7129 break;
7130
7131 rem = tracing_fill_pipe_page(rem, iter);
7132
7133 /* Copy the data into the page, so we can start over. */
7134 ret = trace_seq_to_buffer(&iter->seq,
7135 page_address(spd.pages[i]),
7136 trace_seq_used(&iter->seq));
7137 if (ret < 0) {
7138 __free_page(spd.pages[i]);
7139 break;
7140 }
7141 spd.partial[i].offset = 0;
7142 spd.partial[i].len = trace_seq_used(&iter->seq);
7143
7144 trace_seq_init(&iter->seq);
7145 }
7146
7147 trace_access_unlock(iter->cpu_file);
7148 trace_event_read_unlock();
7149 mutex_unlock(&iter->mutex);
7150
7151 spd.nr_pages = i;
7152
7153 if (i)
7154 ret = splice_to_pipe(pipe, &spd);
7155 else
7156 ret = 0;
7157 out:
7158 splice_shrink_spd(&spd);
7159 return ret;
7160
7161 out_err:
7162 mutex_unlock(&iter->mutex);
7163 goto out;
7164 }
7165
7166 static ssize_t
tracing_entries_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)7167 tracing_entries_read(struct file *filp, char __user *ubuf,
7168 size_t cnt, loff_t *ppos)
7169 {
7170 struct inode *inode = file_inode(filp);
7171 struct trace_array *tr = inode->i_private;
7172 int cpu = tracing_get_cpu(inode);
7173 char buf[64];
7174 int r = 0;
7175 ssize_t ret;
7176
7177 mutex_lock(&trace_types_lock);
7178
7179 if (cpu == RING_BUFFER_ALL_CPUS) {
7180 int cpu, buf_size_same;
7181 unsigned long size;
7182
7183 size = 0;
7184 buf_size_same = 1;
7185 /* check if all cpu sizes are same */
7186 for_each_tracing_cpu(cpu) {
7187 /* fill in the size from first enabled cpu */
7188 if (size == 0)
7189 size = per_cpu_ptr(tr->array_buffer.data, cpu)->entries;
7190 if (size != per_cpu_ptr(tr->array_buffer.data, cpu)->entries) {
7191 buf_size_same = 0;
7192 break;
7193 }
7194 }
7195
7196 if (buf_size_same) {
7197 if (!tr->ring_buffer_expanded)
7198 r = sprintf(buf, "%lu (expanded: %lu)\n",
7199 size >> 10,
7200 trace_buf_size >> 10);
7201 else
7202 r = sprintf(buf, "%lu\n", size >> 10);
7203 } else
7204 r = sprintf(buf, "X\n");
7205 } else
7206 r = sprintf(buf, "%lu\n", per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10);
7207
7208 mutex_unlock(&trace_types_lock);
7209
7210 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7211 return ret;
7212 }
7213
7214 static ssize_t
tracing_entries_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)7215 tracing_entries_write(struct file *filp, const char __user *ubuf,
7216 size_t cnt, loff_t *ppos)
7217 {
7218 struct inode *inode = file_inode(filp);
7219 struct trace_array *tr = inode->i_private;
7220 unsigned long val;
7221 int ret;
7222
7223 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7224 if (ret)
7225 return ret;
7226
7227 /* must have at least 1 entry */
7228 if (!val)
7229 return -EINVAL;
7230
7231 /* value is in KB */
7232 val <<= 10;
7233 ret = tracing_resize_ring_buffer(tr, val, tracing_get_cpu(inode));
7234 if (ret < 0)
7235 return ret;
7236
7237 *ppos += cnt;
7238
7239 return cnt;
7240 }
7241
7242 static ssize_t
tracing_total_entries_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)7243 tracing_total_entries_read(struct file *filp, char __user *ubuf,
7244 size_t cnt, loff_t *ppos)
7245 {
7246 struct trace_array *tr = filp->private_data;
7247 char buf[64];
7248 int r, cpu;
7249 unsigned long size = 0, expanded_size = 0;
7250
7251 mutex_lock(&trace_types_lock);
7252 for_each_tracing_cpu(cpu) {
7253 size += per_cpu_ptr(tr->array_buffer.data, cpu)->entries >> 10;
7254 if (!tr->ring_buffer_expanded)
7255 expanded_size += trace_buf_size >> 10;
7256 }
7257 if (tr->ring_buffer_expanded)
7258 r = sprintf(buf, "%lu\n", size);
7259 else
7260 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
7261 mutex_unlock(&trace_types_lock);
7262
7263 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
7264 }
7265
7266 static ssize_t
tracing_free_buffer_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)7267 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
7268 size_t cnt, loff_t *ppos)
7269 {
7270 /*
7271 * There is no need to read what the user has written, this function
7272 * is just to make sure that there is no error when "echo" is used
7273 */
7274
7275 *ppos += cnt;
7276
7277 return cnt;
7278 }
7279
7280 static int
tracing_free_buffer_release(struct inode * inode,struct file * filp)7281 tracing_free_buffer_release(struct inode *inode, struct file *filp)
7282 {
7283 struct trace_array *tr = inode->i_private;
7284
7285 /* disable tracing ? */
7286 if (tr->trace_flags & TRACE_ITER_STOP_ON_FREE)
7287 tracer_tracing_off(tr);
7288 /* resize the ring buffer to 0 */
7289 tracing_resize_ring_buffer(tr, 0, RING_BUFFER_ALL_CPUS);
7290
7291 trace_array_put(tr);
7292
7293 return 0;
7294 }
7295
7296 #define TRACE_MARKER_MAX_SIZE 4096
7297
7298 static ssize_t
tracing_mark_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * fpos)7299 tracing_mark_write(struct file *filp, const char __user *ubuf,
7300 size_t cnt, loff_t *fpos)
7301 {
7302 struct trace_array *tr = filp->private_data;
7303 struct ring_buffer_event *event;
7304 enum event_trigger_type tt = ETT_NONE;
7305 struct trace_buffer *buffer;
7306 struct print_entry *entry;
7307 int meta_size;
7308 ssize_t written;
7309 size_t size;
7310 int len;
7311
7312 /* Used in tracing_mark_raw_write() as well */
7313 #define FAULTED_STR "<faulted>"
7314 #define FAULTED_SIZE (sizeof(FAULTED_STR) - 1) /* '\0' is already accounted for */
7315
7316 if (tracing_disabled)
7317 return -EINVAL;
7318
7319 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7320 return -EINVAL;
7321
7322 if ((ssize_t)cnt < 0)
7323 return -EINVAL;
7324
7325 if (cnt > TRACE_MARKER_MAX_SIZE)
7326 cnt = TRACE_MARKER_MAX_SIZE;
7327
7328 meta_size = sizeof(*entry) + 2; /* add '\0' and possible '\n' */
7329 again:
7330 size = cnt + meta_size;
7331
7332 /* If less than "<faulted>", then make sure we can still add that */
7333 if (cnt < FAULTED_SIZE)
7334 size += FAULTED_SIZE - cnt;
7335
7336 buffer = tr->array_buffer.buffer;
7337 event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
7338 tracing_gen_ctx());
7339 if (unlikely(!event)) {
7340 /*
7341 * If the size was greater than what was allowed, then
7342 * make it smaller and try again.
7343 */
7344 if (size > ring_buffer_max_event_size(buffer)) {
7345 /* cnt < FAULTED size should never be bigger than max */
7346 if (WARN_ON_ONCE(cnt < FAULTED_SIZE))
7347 return -EBADF;
7348 cnt = ring_buffer_max_event_size(buffer) - meta_size;
7349 /* The above should only happen once */
7350 if (WARN_ON_ONCE(cnt + meta_size == size))
7351 return -EBADF;
7352 goto again;
7353 }
7354
7355 /* Ring buffer disabled, return as if not open for write */
7356 return -EBADF;
7357 }
7358
7359 entry = ring_buffer_event_data(event);
7360 entry->ip = _THIS_IP_;
7361
7362 len = __copy_from_user_inatomic(&entry->buf, ubuf, cnt);
7363 if (len) {
7364 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7365 cnt = FAULTED_SIZE;
7366 written = -EFAULT;
7367 } else
7368 written = cnt;
7369
7370 if (tr->trace_marker_file && !list_empty(&tr->trace_marker_file->triggers)) {
7371 /* do not add \n before testing triggers, but add \0 */
7372 entry->buf[cnt] = '\0';
7373 tt = event_triggers_call(tr->trace_marker_file, buffer, entry, event);
7374 }
7375
7376 if (entry->buf[cnt - 1] != '\n') {
7377 entry->buf[cnt] = '\n';
7378 entry->buf[cnt + 1] = '\0';
7379 } else
7380 entry->buf[cnt] = '\0';
7381
7382 if (static_branch_unlikely(&trace_marker_exports_enabled))
7383 ftrace_exports(event, TRACE_EXPORT_MARKER);
7384 __buffer_unlock_commit(buffer, event);
7385
7386 if (tt)
7387 event_triggers_post_call(tr->trace_marker_file, tt);
7388
7389 return written;
7390 }
7391
7392 static ssize_t
tracing_mark_raw_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * fpos)7393 tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
7394 size_t cnt, loff_t *fpos)
7395 {
7396 struct trace_array *tr = filp->private_data;
7397 struct ring_buffer_event *event;
7398 struct trace_buffer *buffer;
7399 struct raw_data_entry *entry;
7400 ssize_t written;
7401 int size;
7402 int len;
7403
7404 #define FAULT_SIZE_ID (FAULTED_SIZE + sizeof(int))
7405
7406 if (tracing_disabled)
7407 return -EINVAL;
7408
7409 if (!(tr->trace_flags & TRACE_ITER_MARKERS))
7410 return -EINVAL;
7411
7412 /* The marker must at least have a tag id */
7413 if (cnt < sizeof(unsigned int))
7414 return -EINVAL;
7415
7416 size = sizeof(*entry) + cnt;
7417 if (cnt < FAULT_SIZE_ID)
7418 size += FAULT_SIZE_ID - cnt;
7419
7420 buffer = tr->array_buffer.buffer;
7421
7422 if (size > ring_buffer_max_event_size(buffer))
7423 return -EINVAL;
7424
7425 event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
7426 tracing_gen_ctx());
7427 if (!event)
7428 /* Ring buffer disabled, return as if not open for write */
7429 return -EBADF;
7430
7431 entry = ring_buffer_event_data(event);
7432
7433 len = __copy_from_user_inatomic(&entry->id, ubuf, cnt);
7434 if (len) {
7435 entry->id = -1;
7436 memcpy(&entry->buf, FAULTED_STR, FAULTED_SIZE);
7437 written = -EFAULT;
7438 } else
7439 written = cnt;
7440
7441 __buffer_unlock_commit(buffer, event);
7442
7443 return written;
7444 }
7445
tracing_clock_show(struct seq_file * m,void * v)7446 static int tracing_clock_show(struct seq_file *m, void *v)
7447 {
7448 struct trace_array *tr = m->private;
7449 int i;
7450
7451 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
7452 seq_printf(m,
7453 "%s%s%s%s", i ? " " : "",
7454 i == tr->clock_id ? "[" : "", trace_clocks[i].name,
7455 i == tr->clock_id ? "]" : "");
7456 seq_putc(m, '\n');
7457
7458 return 0;
7459 }
7460
tracing_set_clock(struct trace_array * tr,const char * clockstr)7461 int tracing_set_clock(struct trace_array *tr, const char *clockstr)
7462 {
7463 int i;
7464
7465 for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
7466 if (strcmp(trace_clocks[i].name, clockstr) == 0)
7467 break;
7468 }
7469 if (i == ARRAY_SIZE(trace_clocks))
7470 return -EINVAL;
7471
7472 mutex_lock(&trace_types_lock);
7473
7474 tr->clock_id = i;
7475
7476 ring_buffer_set_clock(tr->array_buffer.buffer, trace_clocks[i].func);
7477
7478 /*
7479 * New clock may not be consistent with the previous clock.
7480 * Reset the buffer so that it doesn't have incomparable timestamps.
7481 */
7482 tracing_reset_online_cpus(&tr->array_buffer);
7483
7484 #ifdef CONFIG_TRACER_MAX_TRACE
7485 if (tr->max_buffer.buffer)
7486 ring_buffer_set_clock(tr->max_buffer.buffer, trace_clocks[i].func);
7487 tracing_reset_online_cpus(&tr->max_buffer);
7488 #endif
7489
7490 mutex_unlock(&trace_types_lock);
7491
7492 return 0;
7493 }
7494
tracing_clock_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * fpos)7495 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
7496 size_t cnt, loff_t *fpos)
7497 {
7498 struct seq_file *m = filp->private_data;
7499 struct trace_array *tr = m->private;
7500 char buf[64];
7501 const char *clockstr;
7502 int ret;
7503
7504 if (cnt >= sizeof(buf))
7505 return -EINVAL;
7506
7507 if (copy_from_user(buf, ubuf, cnt))
7508 return -EFAULT;
7509
7510 buf[cnt] = 0;
7511
7512 clockstr = strstrip(buf);
7513
7514 ret = tracing_set_clock(tr, clockstr);
7515 if (ret)
7516 return ret;
7517
7518 *fpos += cnt;
7519
7520 return cnt;
7521 }
7522
tracing_clock_open(struct inode * inode,struct file * file)7523 static int tracing_clock_open(struct inode *inode, struct file *file)
7524 {
7525 struct trace_array *tr = inode->i_private;
7526 int ret;
7527
7528 ret = tracing_check_open_get_tr(tr);
7529 if (ret)
7530 return ret;
7531
7532 ret = single_open(file, tracing_clock_show, inode->i_private);
7533 if (ret < 0)
7534 trace_array_put(tr);
7535
7536 return ret;
7537 }
7538
tracing_time_stamp_mode_show(struct seq_file * m,void * v)7539 static int tracing_time_stamp_mode_show(struct seq_file *m, void *v)
7540 {
7541 struct trace_array *tr = m->private;
7542
7543 mutex_lock(&trace_types_lock);
7544
7545 if (ring_buffer_time_stamp_abs(tr->array_buffer.buffer))
7546 seq_puts(m, "delta [absolute]\n");
7547 else
7548 seq_puts(m, "[delta] absolute\n");
7549
7550 mutex_unlock(&trace_types_lock);
7551
7552 return 0;
7553 }
7554
tracing_time_stamp_mode_open(struct inode * inode,struct file * file)7555 static int tracing_time_stamp_mode_open(struct inode *inode, struct file *file)
7556 {
7557 struct trace_array *tr = inode->i_private;
7558 int ret;
7559
7560 ret = tracing_check_open_get_tr(tr);
7561 if (ret)
7562 return ret;
7563
7564 ret = single_open(file, tracing_time_stamp_mode_show, inode->i_private);
7565 if (ret < 0)
7566 trace_array_put(tr);
7567
7568 return ret;
7569 }
7570
tracing_event_time_stamp(struct trace_buffer * buffer,struct ring_buffer_event * rbe)7571 u64 tracing_event_time_stamp(struct trace_buffer *buffer, struct ring_buffer_event *rbe)
7572 {
7573 if (rbe == this_cpu_read(trace_buffered_event))
7574 return ring_buffer_time_stamp(buffer);
7575
7576 return ring_buffer_event_time_stamp(buffer, rbe);
7577 }
7578
7579 /*
7580 * Set or disable using the per CPU trace_buffer_event when possible.
7581 */
tracing_set_filter_buffering(struct trace_array * tr,bool set)7582 int tracing_set_filter_buffering(struct trace_array *tr, bool set)
7583 {
7584 int ret = 0;
7585
7586 mutex_lock(&trace_types_lock);
7587
7588 if (set && tr->no_filter_buffering_ref++)
7589 goto out;
7590
7591 if (!set) {
7592 if (WARN_ON_ONCE(!tr->no_filter_buffering_ref)) {
7593 ret = -EINVAL;
7594 goto out;
7595 }
7596
7597 --tr->no_filter_buffering_ref;
7598 }
7599 out:
7600 mutex_unlock(&trace_types_lock);
7601
7602 return ret;
7603 }
7604
7605 struct ftrace_buffer_info {
7606 struct trace_iterator iter;
7607 void *spare;
7608 unsigned int spare_cpu;
7609 unsigned int spare_size;
7610 unsigned int read;
7611 };
7612
7613 #ifdef CONFIG_TRACER_SNAPSHOT
tracing_snapshot_open(struct inode * inode,struct file * file)7614 static int tracing_snapshot_open(struct inode *inode, struct file *file)
7615 {
7616 struct trace_array *tr = inode->i_private;
7617 struct trace_iterator *iter;
7618 struct seq_file *m;
7619 int ret;
7620
7621 ret = tracing_check_open_get_tr(tr);
7622 if (ret)
7623 return ret;
7624
7625 if (file->f_mode & FMODE_READ) {
7626 iter = __tracing_open(inode, file, true);
7627 if (IS_ERR(iter))
7628 ret = PTR_ERR(iter);
7629 } else {
7630 /* Writes still need the seq_file to hold the private data */
7631 ret = -ENOMEM;
7632 m = kzalloc(sizeof(*m), GFP_KERNEL);
7633 if (!m)
7634 goto out;
7635 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
7636 if (!iter) {
7637 kfree(m);
7638 goto out;
7639 }
7640 ret = 0;
7641
7642 iter->tr = tr;
7643 iter->array_buffer = &tr->max_buffer;
7644 iter->cpu_file = tracing_get_cpu(inode);
7645 m->private = iter;
7646 file->private_data = m;
7647 }
7648 out:
7649 if (ret < 0)
7650 trace_array_put(tr);
7651
7652 return ret;
7653 }
7654
tracing_swap_cpu_buffer(void * tr)7655 static void tracing_swap_cpu_buffer(void *tr)
7656 {
7657 update_max_tr_single((struct trace_array *)tr, current, smp_processor_id());
7658 }
7659
7660 static ssize_t
tracing_snapshot_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)7661 tracing_snapshot_write(struct file *filp, const char __user *ubuf, size_t cnt,
7662 loff_t *ppos)
7663 {
7664 struct seq_file *m = filp->private_data;
7665 struct trace_iterator *iter = m->private;
7666 struct trace_array *tr = iter->tr;
7667 unsigned long val;
7668 int ret;
7669
7670 ret = tracing_update_buffers(tr);
7671 if (ret < 0)
7672 return ret;
7673
7674 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
7675 if (ret)
7676 return ret;
7677
7678 mutex_lock(&trace_types_lock);
7679
7680 if (tr->current_trace->use_max_tr) {
7681 ret = -EBUSY;
7682 goto out;
7683 }
7684
7685 local_irq_disable();
7686 arch_spin_lock(&tr->max_lock);
7687 if (tr->cond_snapshot)
7688 ret = -EBUSY;
7689 arch_spin_unlock(&tr->max_lock);
7690 local_irq_enable();
7691 if (ret)
7692 goto out;
7693
7694 switch (val) {
7695 case 0:
7696 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7697 ret = -EINVAL;
7698 break;
7699 }
7700 if (tr->allocated_snapshot)
7701 free_snapshot(tr);
7702 break;
7703 case 1:
7704 /* Only allow per-cpu swap if the ring buffer supports it */
7705 #ifndef CONFIG_RING_BUFFER_ALLOW_SWAP
7706 if (iter->cpu_file != RING_BUFFER_ALL_CPUS) {
7707 ret = -EINVAL;
7708 break;
7709 }
7710 #endif
7711 if (tr->allocated_snapshot)
7712 ret = resize_buffer_duplicate_size(&tr->max_buffer,
7713 &tr->array_buffer, iter->cpu_file);
7714 else
7715 ret = tracing_alloc_snapshot_instance(tr);
7716 if (ret < 0)
7717 break;
7718 /* Now, we're going to swap */
7719 if (iter->cpu_file == RING_BUFFER_ALL_CPUS) {
7720 local_irq_disable();
7721 update_max_tr(tr, current, smp_processor_id(), NULL);
7722 local_irq_enable();
7723 } else {
7724 smp_call_function_single(iter->cpu_file, tracing_swap_cpu_buffer,
7725 (void *)tr, 1);
7726 }
7727 break;
7728 default:
7729 if (tr->allocated_snapshot) {
7730 if (iter->cpu_file == RING_BUFFER_ALL_CPUS)
7731 tracing_reset_online_cpus(&tr->max_buffer);
7732 else
7733 tracing_reset_cpu(&tr->max_buffer, iter->cpu_file);
7734 }
7735 break;
7736 }
7737
7738 if (ret >= 0) {
7739 *ppos += cnt;
7740 ret = cnt;
7741 }
7742 out:
7743 mutex_unlock(&trace_types_lock);
7744 return ret;
7745 }
7746
tracing_snapshot_release(struct inode * inode,struct file * file)7747 static int tracing_snapshot_release(struct inode *inode, struct file *file)
7748 {
7749 struct seq_file *m = file->private_data;
7750 int ret;
7751
7752 ret = tracing_release(inode, file);
7753
7754 if (file->f_mode & FMODE_READ)
7755 return ret;
7756
7757 /* If write only, the seq_file is just a stub */
7758 if (m)
7759 kfree(m->private);
7760 kfree(m);
7761
7762 return 0;
7763 }
7764
7765 static int tracing_buffers_open(struct inode *inode, struct file *filp);
7766 static ssize_t tracing_buffers_read(struct file *filp, char __user *ubuf,
7767 size_t count, loff_t *ppos);
7768 static int tracing_buffers_release(struct inode *inode, struct file *file);
7769 static ssize_t tracing_buffers_splice_read(struct file *file, loff_t *ppos,
7770 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
7771
snapshot_raw_open(struct inode * inode,struct file * filp)7772 static int snapshot_raw_open(struct inode *inode, struct file *filp)
7773 {
7774 struct ftrace_buffer_info *info;
7775 int ret;
7776
7777 /* The following checks for tracefs lockdown */
7778 ret = tracing_buffers_open(inode, filp);
7779 if (ret < 0)
7780 return ret;
7781
7782 info = filp->private_data;
7783
7784 if (info->iter.trace->use_max_tr) {
7785 tracing_buffers_release(inode, filp);
7786 return -EBUSY;
7787 }
7788
7789 info->iter.snapshot = true;
7790 info->iter.array_buffer = &info->iter.tr->max_buffer;
7791
7792 return ret;
7793 }
7794
7795 #endif /* CONFIG_TRACER_SNAPSHOT */
7796
7797
7798 static const struct file_operations tracing_thresh_fops = {
7799 .open = tracing_open_generic,
7800 .read = tracing_thresh_read,
7801 .write = tracing_thresh_write,
7802 .llseek = generic_file_llseek,
7803 };
7804
7805 #ifdef CONFIG_TRACER_MAX_TRACE
7806 static const struct file_operations tracing_max_lat_fops = {
7807 .open = tracing_open_generic_tr,
7808 .read = tracing_max_lat_read,
7809 .write = tracing_max_lat_write,
7810 .llseek = generic_file_llseek,
7811 .release = tracing_release_generic_tr,
7812 };
7813 #endif
7814
7815 static const struct file_operations set_tracer_fops = {
7816 .open = tracing_open_generic_tr,
7817 .read = tracing_set_trace_read,
7818 .write = tracing_set_trace_write,
7819 .llseek = generic_file_llseek,
7820 .release = tracing_release_generic_tr,
7821 };
7822
7823 static const struct file_operations tracing_pipe_fops = {
7824 .open = tracing_open_pipe,
7825 .poll = tracing_poll_pipe,
7826 .read = tracing_read_pipe,
7827 .splice_read = tracing_splice_read_pipe,
7828 .release = tracing_release_pipe,
7829 .llseek = no_llseek,
7830 };
7831
7832 static const struct file_operations tracing_entries_fops = {
7833 .open = tracing_open_generic_tr,
7834 .read = tracing_entries_read,
7835 .write = tracing_entries_write,
7836 .llseek = generic_file_llseek,
7837 .release = tracing_release_generic_tr,
7838 };
7839
7840 static const struct file_operations tracing_total_entries_fops = {
7841 .open = tracing_open_generic_tr,
7842 .read = tracing_total_entries_read,
7843 .llseek = generic_file_llseek,
7844 .release = tracing_release_generic_tr,
7845 };
7846
7847 static const struct file_operations tracing_free_buffer_fops = {
7848 .open = tracing_open_generic_tr,
7849 .write = tracing_free_buffer_write,
7850 .release = tracing_free_buffer_release,
7851 };
7852
7853 static const struct file_operations tracing_mark_fops = {
7854 .open = tracing_mark_open,
7855 .write = tracing_mark_write,
7856 .release = tracing_release_generic_tr,
7857 };
7858
7859 static const struct file_operations tracing_mark_raw_fops = {
7860 .open = tracing_mark_open,
7861 .write = tracing_mark_raw_write,
7862 .release = tracing_release_generic_tr,
7863 };
7864
7865 static const struct file_operations trace_clock_fops = {
7866 .open = tracing_clock_open,
7867 .read = seq_read,
7868 .llseek = seq_lseek,
7869 .release = tracing_single_release_tr,
7870 .write = tracing_clock_write,
7871 };
7872
7873 static const struct file_operations trace_time_stamp_mode_fops = {
7874 .open = tracing_time_stamp_mode_open,
7875 .read = seq_read,
7876 .llseek = seq_lseek,
7877 .release = tracing_single_release_tr,
7878 };
7879
7880 #ifdef CONFIG_TRACER_SNAPSHOT
7881 static const struct file_operations snapshot_fops = {
7882 .open = tracing_snapshot_open,
7883 .read = seq_read,
7884 .write = tracing_snapshot_write,
7885 .llseek = tracing_lseek,
7886 .release = tracing_snapshot_release,
7887 };
7888
7889 static const struct file_operations snapshot_raw_fops = {
7890 .open = snapshot_raw_open,
7891 .read = tracing_buffers_read,
7892 .release = tracing_buffers_release,
7893 .splice_read = tracing_buffers_splice_read,
7894 .llseek = no_llseek,
7895 };
7896
7897 #endif /* CONFIG_TRACER_SNAPSHOT */
7898
7899 /*
7900 * trace_min_max_write - Write a u64 value to a trace_min_max_param struct
7901 * @filp: The active open file structure
7902 * @ubuf: The userspace provided buffer to read value into
7903 * @cnt: The maximum number of bytes to read
7904 * @ppos: The current "file" position
7905 *
7906 * This function implements the write interface for a struct trace_min_max_param.
7907 * The filp->private_data must point to a trace_min_max_param structure that
7908 * defines where to write the value, the min and the max acceptable values,
7909 * and a lock to protect the write.
7910 */
7911 static ssize_t
trace_min_max_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)7912 trace_min_max_write(struct file *filp, const char __user *ubuf, size_t cnt, loff_t *ppos)
7913 {
7914 struct trace_min_max_param *param = filp->private_data;
7915 u64 val;
7916 int err;
7917
7918 if (!param)
7919 return -EFAULT;
7920
7921 err = kstrtoull_from_user(ubuf, cnt, 10, &val);
7922 if (err)
7923 return err;
7924
7925 if (param->lock)
7926 mutex_lock(param->lock);
7927
7928 if (param->min && val < *param->min)
7929 err = -EINVAL;
7930
7931 if (param->max && val > *param->max)
7932 err = -EINVAL;
7933
7934 if (!err)
7935 *param->val = val;
7936
7937 if (param->lock)
7938 mutex_unlock(param->lock);
7939
7940 if (err)
7941 return err;
7942
7943 return cnt;
7944 }
7945
7946 /*
7947 * trace_min_max_read - Read a u64 value from a trace_min_max_param struct
7948 * @filp: The active open file structure
7949 * @ubuf: The userspace provided buffer to read value into
7950 * @cnt: The maximum number of bytes to read
7951 * @ppos: The current "file" position
7952 *
7953 * This function implements the read interface for a struct trace_min_max_param.
7954 * The filp->private_data must point to a trace_min_max_param struct with valid
7955 * data.
7956 */
7957 static ssize_t
trace_min_max_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)7958 trace_min_max_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
7959 {
7960 struct trace_min_max_param *param = filp->private_data;
7961 char buf[U64_STR_SIZE];
7962 int len;
7963 u64 val;
7964
7965 if (!param)
7966 return -EFAULT;
7967
7968 val = *param->val;
7969
7970 if (cnt > sizeof(buf))
7971 cnt = sizeof(buf);
7972
7973 len = snprintf(buf, sizeof(buf), "%llu\n", val);
7974
7975 return simple_read_from_buffer(ubuf, cnt, ppos, buf, len);
7976 }
7977
7978 const struct file_operations trace_min_max_fops = {
7979 .open = tracing_open_generic,
7980 .read = trace_min_max_read,
7981 .write = trace_min_max_write,
7982 };
7983
7984 #define TRACING_LOG_ERRS_MAX 8
7985 #define TRACING_LOG_LOC_MAX 128
7986
7987 #define CMD_PREFIX " Command: "
7988
7989 struct err_info {
7990 const char **errs; /* ptr to loc-specific array of err strings */
7991 u8 type; /* index into errs -> specific err string */
7992 u16 pos; /* caret position */
7993 u64 ts;
7994 };
7995
7996 struct tracing_log_err {
7997 struct list_head list;
7998 struct err_info info;
7999 char loc[TRACING_LOG_LOC_MAX]; /* err location */
8000 char *cmd; /* what caused err */
8001 };
8002
8003 static DEFINE_MUTEX(tracing_err_log_lock);
8004
alloc_tracing_log_err(int len)8005 static struct tracing_log_err *alloc_tracing_log_err(int len)
8006 {
8007 struct tracing_log_err *err;
8008
8009 err = kzalloc(sizeof(*err), GFP_KERNEL);
8010 if (!err)
8011 return ERR_PTR(-ENOMEM);
8012
8013 err->cmd = kzalloc(len, GFP_KERNEL);
8014 if (!err->cmd) {
8015 kfree(err);
8016 return ERR_PTR(-ENOMEM);
8017 }
8018
8019 return err;
8020 }
8021
free_tracing_log_err(struct tracing_log_err * err)8022 static void free_tracing_log_err(struct tracing_log_err *err)
8023 {
8024 kfree(err->cmd);
8025 kfree(err);
8026 }
8027
get_tracing_log_err(struct trace_array * tr,int len)8028 static struct tracing_log_err *get_tracing_log_err(struct trace_array *tr,
8029 int len)
8030 {
8031 struct tracing_log_err *err;
8032 char *cmd;
8033
8034 if (tr->n_err_log_entries < TRACING_LOG_ERRS_MAX) {
8035 err = alloc_tracing_log_err(len);
8036 if (PTR_ERR(err) != -ENOMEM)
8037 tr->n_err_log_entries++;
8038
8039 return err;
8040 }
8041 cmd = kzalloc(len, GFP_KERNEL);
8042 if (!cmd)
8043 return ERR_PTR(-ENOMEM);
8044 err = list_first_entry(&tr->err_log, struct tracing_log_err, list);
8045 kfree(err->cmd);
8046 err->cmd = cmd;
8047 list_del(&err->list);
8048
8049 return err;
8050 }
8051
8052 /**
8053 * err_pos - find the position of a string within a command for error careting
8054 * @cmd: The tracing command that caused the error
8055 * @str: The string to position the caret at within @cmd
8056 *
8057 * Finds the position of the first occurrence of @str within @cmd. The
8058 * return value can be passed to tracing_log_err() for caret placement
8059 * within @cmd.
8060 *
8061 * Returns the index within @cmd of the first occurrence of @str or 0
8062 * if @str was not found.
8063 */
err_pos(char * cmd,const char * str)8064 unsigned int err_pos(char *cmd, const char *str)
8065 {
8066 char *found;
8067
8068 if (WARN_ON(!strlen(cmd)))
8069 return 0;
8070
8071 found = strstr(cmd, str);
8072 if (found)
8073 return found - cmd;
8074
8075 return 0;
8076 }
8077
8078 /**
8079 * tracing_log_err - write an error to the tracing error log
8080 * @tr: The associated trace array for the error (NULL for top level array)
8081 * @loc: A string describing where the error occurred
8082 * @cmd: The tracing command that caused the error
8083 * @errs: The array of loc-specific static error strings
8084 * @type: The index into errs[], which produces the specific static err string
8085 * @pos: The position the caret should be placed in the cmd
8086 *
8087 * Writes an error into tracing/error_log of the form:
8088 *
8089 * <loc>: error: <text>
8090 * Command: <cmd>
8091 * ^
8092 *
8093 * tracing/error_log is a small log file containing the last
8094 * TRACING_LOG_ERRS_MAX errors (8). Memory for errors isn't allocated
8095 * unless there has been a tracing error, and the error log can be
8096 * cleared and have its memory freed by writing the empty string in
8097 * truncation mode to it i.e. echo > tracing/error_log.
8098 *
8099 * NOTE: the @errs array along with the @type param are used to
8100 * produce a static error string - this string is not copied and saved
8101 * when the error is logged - only a pointer to it is saved. See
8102 * existing callers for examples of how static strings are typically
8103 * defined for use with tracing_log_err().
8104 */
tracing_log_err(struct trace_array * tr,const char * loc,const char * cmd,const char ** errs,u8 type,u16 pos)8105 void tracing_log_err(struct trace_array *tr,
8106 const char *loc, const char *cmd,
8107 const char **errs, u8 type, u16 pos)
8108 {
8109 struct tracing_log_err *err;
8110 int len = 0;
8111
8112 if (!tr)
8113 tr = &global_trace;
8114
8115 len += sizeof(CMD_PREFIX) + 2 * sizeof("\n") + strlen(cmd) + 1;
8116
8117 mutex_lock(&tracing_err_log_lock);
8118 err = get_tracing_log_err(tr, len);
8119 if (PTR_ERR(err) == -ENOMEM) {
8120 mutex_unlock(&tracing_err_log_lock);
8121 return;
8122 }
8123
8124 snprintf(err->loc, TRACING_LOG_LOC_MAX, "%s: error: ", loc);
8125 snprintf(err->cmd, len, "\n" CMD_PREFIX "%s\n", cmd);
8126
8127 err->info.errs = errs;
8128 err->info.type = type;
8129 err->info.pos = pos;
8130 err->info.ts = local_clock();
8131
8132 list_add_tail(&err->list, &tr->err_log);
8133 mutex_unlock(&tracing_err_log_lock);
8134 }
8135
clear_tracing_err_log(struct trace_array * tr)8136 static void clear_tracing_err_log(struct trace_array *tr)
8137 {
8138 struct tracing_log_err *err, *next;
8139
8140 mutex_lock(&tracing_err_log_lock);
8141 list_for_each_entry_safe(err, next, &tr->err_log, list) {
8142 list_del(&err->list);
8143 free_tracing_log_err(err);
8144 }
8145
8146 tr->n_err_log_entries = 0;
8147 mutex_unlock(&tracing_err_log_lock);
8148 }
8149
tracing_err_log_seq_start(struct seq_file * m,loff_t * pos)8150 static void *tracing_err_log_seq_start(struct seq_file *m, loff_t *pos)
8151 {
8152 struct trace_array *tr = m->private;
8153
8154 mutex_lock(&tracing_err_log_lock);
8155
8156 return seq_list_start(&tr->err_log, *pos);
8157 }
8158
tracing_err_log_seq_next(struct seq_file * m,void * v,loff_t * pos)8159 static void *tracing_err_log_seq_next(struct seq_file *m, void *v, loff_t *pos)
8160 {
8161 struct trace_array *tr = m->private;
8162
8163 return seq_list_next(v, &tr->err_log, pos);
8164 }
8165
tracing_err_log_seq_stop(struct seq_file * m,void * v)8166 static void tracing_err_log_seq_stop(struct seq_file *m, void *v)
8167 {
8168 mutex_unlock(&tracing_err_log_lock);
8169 }
8170
tracing_err_log_show_pos(struct seq_file * m,u16 pos)8171 static void tracing_err_log_show_pos(struct seq_file *m, u16 pos)
8172 {
8173 u16 i;
8174
8175 for (i = 0; i < sizeof(CMD_PREFIX) - 1; i++)
8176 seq_putc(m, ' ');
8177 for (i = 0; i < pos; i++)
8178 seq_putc(m, ' ');
8179 seq_puts(m, "^\n");
8180 }
8181
tracing_err_log_seq_show(struct seq_file * m,void * v)8182 static int tracing_err_log_seq_show(struct seq_file *m, void *v)
8183 {
8184 struct tracing_log_err *err = v;
8185
8186 if (err) {
8187 const char *err_text = err->info.errs[err->info.type];
8188 u64 sec = err->info.ts;
8189 u32 nsec;
8190
8191 nsec = do_div(sec, NSEC_PER_SEC);
8192 seq_printf(m, "[%5llu.%06u] %s%s", sec, nsec / 1000,
8193 err->loc, err_text);
8194 seq_printf(m, "%s", err->cmd);
8195 tracing_err_log_show_pos(m, err->info.pos);
8196 }
8197
8198 return 0;
8199 }
8200
8201 static const struct seq_operations tracing_err_log_seq_ops = {
8202 .start = tracing_err_log_seq_start,
8203 .next = tracing_err_log_seq_next,
8204 .stop = tracing_err_log_seq_stop,
8205 .show = tracing_err_log_seq_show
8206 };
8207
tracing_err_log_open(struct inode * inode,struct file * file)8208 static int tracing_err_log_open(struct inode *inode, struct file *file)
8209 {
8210 struct trace_array *tr = inode->i_private;
8211 int ret = 0;
8212
8213 ret = tracing_check_open_get_tr(tr);
8214 if (ret)
8215 return ret;
8216
8217 /* If this file was opened for write, then erase contents */
8218 if ((file->f_mode & FMODE_WRITE) && (file->f_flags & O_TRUNC))
8219 clear_tracing_err_log(tr);
8220
8221 if (file->f_mode & FMODE_READ) {
8222 ret = seq_open(file, &tracing_err_log_seq_ops);
8223 if (!ret) {
8224 struct seq_file *m = file->private_data;
8225 m->private = tr;
8226 } else {
8227 trace_array_put(tr);
8228 }
8229 }
8230 return ret;
8231 }
8232
tracing_err_log_write(struct file * file,const char __user * buffer,size_t count,loff_t * ppos)8233 static ssize_t tracing_err_log_write(struct file *file,
8234 const char __user *buffer,
8235 size_t count, loff_t *ppos)
8236 {
8237 return count;
8238 }
8239
tracing_err_log_release(struct inode * inode,struct file * file)8240 static int tracing_err_log_release(struct inode *inode, struct file *file)
8241 {
8242 struct trace_array *tr = inode->i_private;
8243
8244 trace_array_put(tr);
8245
8246 if (file->f_mode & FMODE_READ)
8247 seq_release(inode, file);
8248
8249 return 0;
8250 }
8251
8252 static const struct file_operations tracing_err_log_fops = {
8253 .open = tracing_err_log_open,
8254 .write = tracing_err_log_write,
8255 .read = seq_read,
8256 .llseek = tracing_lseek,
8257 .release = tracing_err_log_release,
8258 };
8259
tracing_buffers_open(struct inode * inode,struct file * filp)8260 static int tracing_buffers_open(struct inode *inode, struct file *filp)
8261 {
8262 struct trace_array *tr = inode->i_private;
8263 struct ftrace_buffer_info *info;
8264 int ret;
8265
8266 ret = tracing_check_open_get_tr(tr);
8267 if (ret)
8268 return ret;
8269
8270 info = kvzalloc(sizeof(*info), GFP_KERNEL);
8271 if (!info) {
8272 trace_array_put(tr);
8273 return -ENOMEM;
8274 }
8275
8276 mutex_lock(&trace_types_lock);
8277
8278 info->iter.tr = tr;
8279 info->iter.cpu_file = tracing_get_cpu(inode);
8280 info->iter.trace = tr->current_trace;
8281 info->iter.array_buffer = &tr->array_buffer;
8282 info->spare = NULL;
8283 /* Force reading ring buffer for first read */
8284 info->read = (unsigned int)-1;
8285
8286 filp->private_data = info;
8287
8288 tr->trace_ref++;
8289
8290 mutex_unlock(&trace_types_lock);
8291
8292 ret = nonseekable_open(inode, filp);
8293 if (ret < 0)
8294 trace_array_put(tr);
8295
8296 return ret;
8297 }
8298
8299 static __poll_t
tracing_buffers_poll(struct file * filp,poll_table * poll_table)8300 tracing_buffers_poll(struct file *filp, poll_table *poll_table)
8301 {
8302 struct ftrace_buffer_info *info = filp->private_data;
8303 struct trace_iterator *iter = &info->iter;
8304
8305 return trace_poll(iter, filp, poll_table);
8306 }
8307
8308 static ssize_t
tracing_buffers_read(struct file * filp,char __user * ubuf,size_t count,loff_t * ppos)8309 tracing_buffers_read(struct file *filp, char __user *ubuf,
8310 size_t count, loff_t *ppos)
8311 {
8312 struct ftrace_buffer_info *info = filp->private_data;
8313 struct trace_iterator *iter = &info->iter;
8314 void *trace_data;
8315 int page_size;
8316 ssize_t ret = 0;
8317 ssize_t size;
8318
8319 if (!count)
8320 return 0;
8321
8322 #ifdef CONFIG_TRACER_MAX_TRACE
8323 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8324 return -EBUSY;
8325 #endif
8326
8327 page_size = ring_buffer_subbuf_size_get(iter->array_buffer->buffer);
8328
8329 /* Make sure the spare matches the current sub buffer size */
8330 if (info->spare) {
8331 if (page_size != info->spare_size) {
8332 ring_buffer_free_read_page(iter->array_buffer->buffer,
8333 info->spare_cpu, info->spare);
8334 info->spare = NULL;
8335 }
8336 }
8337
8338 if (!info->spare) {
8339 info->spare = ring_buffer_alloc_read_page(iter->array_buffer->buffer,
8340 iter->cpu_file);
8341 if (IS_ERR(info->spare)) {
8342 ret = PTR_ERR(info->spare);
8343 info->spare = NULL;
8344 } else {
8345 info->spare_cpu = iter->cpu_file;
8346 info->spare_size = page_size;
8347 }
8348 }
8349 if (!info->spare)
8350 return ret;
8351
8352 /* Do we have previous read data to read? */
8353 if (info->read < page_size)
8354 goto read;
8355
8356 again:
8357 trace_access_lock(iter->cpu_file);
8358 ret = ring_buffer_read_page(iter->array_buffer->buffer,
8359 info->spare,
8360 count,
8361 iter->cpu_file, 0);
8362 trace_access_unlock(iter->cpu_file);
8363
8364 if (ret < 0) {
8365 if (trace_empty(iter)) {
8366 if ((filp->f_flags & O_NONBLOCK))
8367 return -EAGAIN;
8368
8369 ret = wait_on_pipe(iter, 0);
8370 if (ret)
8371 return ret;
8372
8373 goto again;
8374 }
8375 return 0;
8376 }
8377
8378 info->read = 0;
8379 read:
8380 size = page_size - info->read;
8381 if (size > count)
8382 size = count;
8383 trace_data = ring_buffer_read_page_data(info->spare);
8384 ret = copy_to_user(ubuf, trace_data + info->read, size);
8385 if (ret == size)
8386 return -EFAULT;
8387
8388 size -= ret;
8389
8390 *ppos += size;
8391 info->read += size;
8392
8393 return size;
8394 }
8395
tracing_buffers_flush(struct file * file,fl_owner_t id)8396 static int tracing_buffers_flush(struct file *file, fl_owner_t id)
8397 {
8398 struct ftrace_buffer_info *info = file->private_data;
8399 struct trace_iterator *iter = &info->iter;
8400
8401 iter->wait_index++;
8402 /* Make sure the waiters see the new wait_index */
8403 smp_wmb();
8404
8405 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8406
8407 return 0;
8408 }
8409
tracing_buffers_release(struct inode * inode,struct file * file)8410 static int tracing_buffers_release(struct inode *inode, struct file *file)
8411 {
8412 struct ftrace_buffer_info *info = file->private_data;
8413 struct trace_iterator *iter = &info->iter;
8414
8415 mutex_lock(&trace_types_lock);
8416
8417 iter->tr->trace_ref--;
8418
8419 __trace_array_put(iter->tr);
8420
8421 if (info->spare)
8422 ring_buffer_free_read_page(iter->array_buffer->buffer,
8423 info->spare_cpu, info->spare);
8424 kvfree(info);
8425
8426 mutex_unlock(&trace_types_lock);
8427
8428 return 0;
8429 }
8430
8431 struct buffer_ref {
8432 struct trace_buffer *buffer;
8433 void *page;
8434 int cpu;
8435 refcount_t refcount;
8436 };
8437
buffer_ref_release(struct buffer_ref * ref)8438 static void buffer_ref_release(struct buffer_ref *ref)
8439 {
8440 if (!refcount_dec_and_test(&ref->refcount))
8441 return;
8442 ring_buffer_free_read_page(ref->buffer, ref->cpu, ref->page);
8443 kfree(ref);
8444 }
8445
buffer_pipe_buf_release(struct pipe_inode_info * pipe,struct pipe_buffer * buf)8446 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
8447 struct pipe_buffer *buf)
8448 {
8449 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8450
8451 buffer_ref_release(ref);
8452 buf->private = 0;
8453 }
8454
buffer_pipe_buf_get(struct pipe_inode_info * pipe,struct pipe_buffer * buf)8455 static bool buffer_pipe_buf_get(struct pipe_inode_info *pipe,
8456 struct pipe_buffer *buf)
8457 {
8458 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
8459
8460 if (refcount_read(&ref->refcount) > INT_MAX/2)
8461 return false;
8462
8463 refcount_inc(&ref->refcount);
8464 return true;
8465 }
8466
8467 /* Pipe buffer operations for a buffer. */
8468 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
8469 .release = buffer_pipe_buf_release,
8470 .get = buffer_pipe_buf_get,
8471 };
8472
8473 /*
8474 * Callback from splice_to_pipe(), if we need to release some pages
8475 * at the end of the spd in case we error'ed out in filling the pipe.
8476 */
buffer_spd_release(struct splice_pipe_desc * spd,unsigned int i)8477 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
8478 {
8479 struct buffer_ref *ref =
8480 (struct buffer_ref *)spd->partial[i].private;
8481
8482 buffer_ref_release(ref);
8483 spd->partial[i].private = 0;
8484 }
8485
8486 static ssize_t
tracing_buffers_splice_read(struct file * file,loff_t * ppos,struct pipe_inode_info * pipe,size_t len,unsigned int flags)8487 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
8488 struct pipe_inode_info *pipe, size_t len,
8489 unsigned int flags)
8490 {
8491 struct ftrace_buffer_info *info = file->private_data;
8492 struct trace_iterator *iter = &info->iter;
8493 struct partial_page partial_def[PIPE_DEF_BUFFERS];
8494 struct page *pages_def[PIPE_DEF_BUFFERS];
8495 struct splice_pipe_desc spd = {
8496 .pages = pages_def,
8497 .partial = partial_def,
8498 .nr_pages_max = PIPE_DEF_BUFFERS,
8499 .ops = &buffer_pipe_buf_ops,
8500 .spd_release = buffer_spd_release,
8501 };
8502 struct buffer_ref *ref;
8503 int page_size;
8504 int entries, i;
8505 ssize_t ret = 0;
8506
8507 #ifdef CONFIG_TRACER_MAX_TRACE
8508 if (iter->snapshot && iter->tr->current_trace->use_max_tr)
8509 return -EBUSY;
8510 #endif
8511
8512 page_size = ring_buffer_subbuf_size_get(iter->array_buffer->buffer);
8513 if (*ppos & (page_size - 1))
8514 return -EINVAL;
8515
8516 if (len & (page_size - 1)) {
8517 if (len < page_size)
8518 return -EINVAL;
8519 len &= (~(page_size - 1));
8520 }
8521
8522 if (splice_grow_spd(pipe, &spd))
8523 return -ENOMEM;
8524
8525 again:
8526 trace_access_lock(iter->cpu_file);
8527 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8528
8529 for (i = 0; i < spd.nr_pages_max && len && entries; i++, len -= page_size) {
8530 struct page *page;
8531 int r;
8532
8533 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
8534 if (!ref) {
8535 ret = -ENOMEM;
8536 break;
8537 }
8538
8539 refcount_set(&ref->refcount, 1);
8540 ref->buffer = iter->array_buffer->buffer;
8541 ref->page = ring_buffer_alloc_read_page(ref->buffer, iter->cpu_file);
8542 if (IS_ERR(ref->page)) {
8543 ret = PTR_ERR(ref->page);
8544 ref->page = NULL;
8545 kfree(ref);
8546 break;
8547 }
8548 ref->cpu = iter->cpu_file;
8549
8550 r = ring_buffer_read_page(ref->buffer, ref->page,
8551 len, iter->cpu_file, 1);
8552 if (r < 0) {
8553 ring_buffer_free_read_page(ref->buffer, ref->cpu,
8554 ref->page);
8555 kfree(ref);
8556 break;
8557 }
8558
8559 page = virt_to_page(ring_buffer_read_page_data(ref->page));
8560
8561 spd.pages[i] = page;
8562 spd.partial[i].len = page_size;
8563 spd.partial[i].offset = 0;
8564 spd.partial[i].private = (unsigned long)ref;
8565 spd.nr_pages++;
8566 *ppos += page_size;
8567
8568 entries = ring_buffer_entries_cpu(iter->array_buffer->buffer, iter->cpu_file);
8569 }
8570
8571 trace_access_unlock(iter->cpu_file);
8572 spd.nr_pages = i;
8573
8574 /* did we read anything? */
8575 if (!spd.nr_pages) {
8576 long wait_index;
8577
8578 if (ret)
8579 goto out;
8580
8581 ret = -EAGAIN;
8582 if ((file->f_flags & O_NONBLOCK) || (flags & SPLICE_F_NONBLOCK))
8583 goto out;
8584
8585 wait_index = READ_ONCE(iter->wait_index);
8586
8587 ret = wait_on_pipe(iter, iter->snapshot ? 0 : iter->tr->buffer_percent);
8588 if (ret)
8589 goto out;
8590
8591 /* No need to wait after waking up when tracing is off */
8592 if (!tracer_tracing_is_on(iter->tr))
8593 goto out;
8594
8595 /* Make sure we see the new wait_index */
8596 smp_rmb();
8597 if (wait_index != iter->wait_index)
8598 goto out;
8599
8600 goto again;
8601 }
8602
8603 ret = splice_to_pipe(pipe, &spd);
8604 out:
8605 splice_shrink_spd(&spd);
8606
8607 return ret;
8608 }
8609
8610 /* An ioctl call with cmd 0 to the ring buffer file will wake up all waiters */
tracing_buffers_ioctl(struct file * file,unsigned int cmd,unsigned long arg)8611 static long tracing_buffers_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
8612 {
8613 struct ftrace_buffer_info *info = file->private_data;
8614 struct trace_iterator *iter = &info->iter;
8615
8616 if (cmd)
8617 return -ENOIOCTLCMD;
8618
8619 mutex_lock(&trace_types_lock);
8620
8621 iter->wait_index++;
8622 /* Make sure the waiters see the new wait_index */
8623 smp_wmb();
8624
8625 ring_buffer_wake_waiters(iter->array_buffer->buffer, iter->cpu_file);
8626
8627 mutex_unlock(&trace_types_lock);
8628 return 0;
8629 }
8630
8631 static const struct file_operations tracing_buffers_fops = {
8632 .open = tracing_buffers_open,
8633 .read = tracing_buffers_read,
8634 .poll = tracing_buffers_poll,
8635 .release = tracing_buffers_release,
8636 .flush = tracing_buffers_flush,
8637 .splice_read = tracing_buffers_splice_read,
8638 .unlocked_ioctl = tracing_buffers_ioctl,
8639 .llseek = no_llseek,
8640 };
8641
8642 static ssize_t
tracing_stats_read(struct file * filp,char __user * ubuf,size_t count,loff_t * ppos)8643 tracing_stats_read(struct file *filp, char __user *ubuf,
8644 size_t count, loff_t *ppos)
8645 {
8646 struct inode *inode = file_inode(filp);
8647 struct trace_array *tr = inode->i_private;
8648 struct array_buffer *trace_buf = &tr->array_buffer;
8649 int cpu = tracing_get_cpu(inode);
8650 struct trace_seq *s;
8651 unsigned long cnt;
8652 unsigned long long t;
8653 unsigned long usec_rem;
8654
8655 s = kmalloc(sizeof(*s), GFP_KERNEL);
8656 if (!s)
8657 return -ENOMEM;
8658
8659 trace_seq_init(s);
8660
8661 cnt = ring_buffer_entries_cpu(trace_buf->buffer, cpu);
8662 trace_seq_printf(s, "entries: %ld\n", cnt);
8663
8664 cnt = ring_buffer_overrun_cpu(trace_buf->buffer, cpu);
8665 trace_seq_printf(s, "overrun: %ld\n", cnt);
8666
8667 cnt = ring_buffer_commit_overrun_cpu(trace_buf->buffer, cpu);
8668 trace_seq_printf(s, "commit overrun: %ld\n", cnt);
8669
8670 cnt = ring_buffer_bytes_cpu(trace_buf->buffer, cpu);
8671 trace_seq_printf(s, "bytes: %ld\n", cnt);
8672
8673 if (trace_clocks[tr->clock_id].in_ns) {
8674 /* local or global for trace_clock */
8675 t = ns2usecs(ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8676 usec_rem = do_div(t, USEC_PER_SEC);
8677 trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n",
8678 t, usec_rem);
8679
8680 t = ns2usecs(ring_buffer_time_stamp(trace_buf->buffer));
8681 usec_rem = do_div(t, USEC_PER_SEC);
8682 trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
8683 } else {
8684 /* counter or tsc mode for trace_clock */
8685 trace_seq_printf(s, "oldest event ts: %llu\n",
8686 ring_buffer_oldest_event_ts(trace_buf->buffer, cpu));
8687
8688 trace_seq_printf(s, "now ts: %llu\n",
8689 ring_buffer_time_stamp(trace_buf->buffer));
8690 }
8691
8692 cnt = ring_buffer_dropped_events_cpu(trace_buf->buffer, cpu);
8693 trace_seq_printf(s, "dropped events: %ld\n", cnt);
8694
8695 cnt = ring_buffer_read_events_cpu(trace_buf->buffer, cpu);
8696 trace_seq_printf(s, "read events: %ld\n", cnt);
8697
8698 count = simple_read_from_buffer(ubuf, count, ppos,
8699 s->buffer, trace_seq_used(s));
8700
8701 kfree(s);
8702
8703 return count;
8704 }
8705
8706 static const struct file_operations tracing_stats_fops = {
8707 .open = tracing_open_generic_tr,
8708 .read = tracing_stats_read,
8709 .llseek = generic_file_llseek,
8710 .release = tracing_release_generic_tr,
8711 };
8712
8713 #ifdef CONFIG_DYNAMIC_FTRACE
8714
8715 static ssize_t
tracing_read_dyn_info(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)8716 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
8717 size_t cnt, loff_t *ppos)
8718 {
8719 ssize_t ret;
8720 char *buf;
8721 int r;
8722
8723 /* 256 should be plenty to hold the amount needed */
8724 buf = kmalloc(256, GFP_KERNEL);
8725 if (!buf)
8726 return -ENOMEM;
8727
8728 r = scnprintf(buf, 256, "%ld pages:%ld groups: %ld\n",
8729 ftrace_update_tot_cnt,
8730 ftrace_number_of_pages,
8731 ftrace_number_of_groups);
8732
8733 ret = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
8734 kfree(buf);
8735 return ret;
8736 }
8737
8738 static const struct file_operations tracing_dyn_info_fops = {
8739 .open = tracing_open_generic,
8740 .read = tracing_read_dyn_info,
8741 .llseek = generic_file_llseek,
8742 };
8743 #endif /* CONFIG_DYNAMIC_FTRACE */
8744
8745 #if defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE)
8746 static void
ftrace_snapshot(unsigned long ip,unsigned long parent_ip,struct trace_array * tr,struct ftrace_probe_ops * ops,void * data)8747 ftrace_snapshot(unsigned long ip, unsigned long parent_ip,
8748 struct trace_array *tr, struct ftrace_probe_ops *ops,
8749 void *data)
8750 {
8751 tracing_snapshot_instance(tr);
8752 }
8753
8754 static void
ftrace_count_snapshot(unsigned long ip,unsigned long parent_ip,struct trace_array * tr,struct ftrace_probe_ops * ops,void * data)8755 ftrace_count_snapshot(unsigned long ip, unsigned long parent_ip,
8756 struct trace_array *tr, struct ftrace_probe_ops *ops,
8757 void *data)
8758 {
8759 struct ftrace_func_mapper *mapper = data;
8760 long *count = NULL;
8761
8762 if (mapper)
8763 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8764
8765 if (count) {
8766
8767 if (*count <= 0)
8768 return;
8769
8770 (*count)--;
8771 }
8772
8773 tracing_snapshot_instance(tr);
8774 }
8775
8776 static int
ftrace_snapshot_print(struct seq_file * m,unsigned long ip,struct ftrace_probe_ops * ops,void * data)8777 ftrace_snapshot_print(struct seq_file *m, unsigned long ip,
8778 struct ftrace_probe_ops *ops, void *data)
8779 {
8780 struct ftrace_func_mapper *mapper = data;
8781 long *count = NULL;
8782
8783 seq_printf(m, "%ps:", (void *)ip);
8784
8785 seq_puts(m, "snapshot");
8786
8787 if (mapper)
8788 count = (long *)ftrace_func_mapper_find_ip(mapper, ip);
8789
8790 if (count)
8791 seq_printf(m, ":count=%ld\n", *count);
8792 else
8793 seq_puts(m, ":unlimited\n");
8794
8795 return 0;
8796 }
8797
8798 static int
ftrace_snapshot_init(struct ftrace_probe_ops * ops,struct trace_array * tr,unsigned long ip,void * init_data,void ** data)8799 ftrace_snapshot_init(struct ftrace_probe_ops *ops, struct trace_array *tr,
8800 unsigned long ip, void *init_data, void **data)
8801 {
8802 struct ftrace_func_mapper *mapper = *data;
8803
8804 if (!mapper) {
8805 mapper = allocate_ftrace_func_mapper();
8806 if (!mapper)
8807 return -ENOMEM;
8808 *data = mapper;
8809 }
8810
8811 return ftrace_func_mapper_add_ip(mapper, ip, init_data);
8812 }
8813
8814 static void
ftrace_snapshot_free(struct ftrace_probe_ops * ops,struct trace_array * tr,unsigned long ip,void * data)8815 ftrace_snapshot_free(struct ftrace_probe_ops *ops, struct trace_array *tr,
8816 unsigned long ip, void *data)
8817 {
8818 struct ftrace_func_mapper *mapper = data;
8819
8820 if (!ip) {
8821 if (!mapper)
8822 return;
8823 free_ftrace_func_mapper(mapper, NULL);
8824 return;
8825 }
8826
8827 ftrace_func_mapper_remove_ip(mapper, ip);
8828 }
8829
8830 static struct ftrace_probe_ops snapshot_probe_ops = {
8831 .func = ftrace_snapshot,
8832 .print = ftrace_snapshot_print,
8833 };
8834
8835 static struct ftrace_probe_ops snapshot_count_probe_ops = {
8836 .func = ftrace_count_snapshot,
8837 .print = ftrace_snapshot_print,
8838 .init = ftrace_snapshot_init,
8839 .free = ftrace_snapshot_free,
8840 };
8841
8842 static int
ftrace_trace_snapshot_callback(struct trace_array * tr,struct ftrace_hash * hash,char * glob,char * cmd,char * param,int enable)8843 ftrace_trace_snapshot_callback(struct trace_array *tr, struct ftrace_hash *hash,
8844 char *glob, char *cmd, char *param, int enable)
8845 {
8846 struct ftrace_probe_ops *ops;
8847 void *count = (void *)-1;
8848 char *number;
8849 int ret;
8850
8851 if (!tr)
8852 return -ENODEV;
8853
8854 /* hash funcs only work with set_ftrace_filter */
8855 if (!enable)
8856 return -EINVAL;
8857
8858 ops = param ? &snapshot_count_probe_ops : &snapshot_probe_ops;
8859
8860 if (glob[0] == '!')
8861 return unregister_ftrace_function_probe_func(glob+1, tr, ops);
8862
8863 if (!param)
8864 goto out_reg;
8865
8866 number = strsep(¶m, ":");
8867
8868 if (!strlen(number))
8869 goto out_reg;
8870
8871 /*
8872 * We use the callback data field (which is a pointer)
8873 * as our counter.
8874 */
8875 ret = kstrtoul(number, 0, (unsigned long *)&count);
8876 if (ret)
8877 return ret;
8878
8879 out_reg:
8880 ret = tracing_alloc_snapshot_instance(tr);
8881 if (ret < 0)
8882 goto out;
8883
8884 ret = register_ftrace_function_probe(glob, tr, ops, count);
8885
8886 out:
8887 return ret < 0 ? ret : 0;
8888 }
8889
8890 static struct ftrace_func_command ftrace_snapshot_cmd = {
8891 .name = "snapshot",
8892 .func = ftrace_trace_snapshot_callback,
8893 };
8894
register_snapshot_cmd(void)8895 static __init int register_snapshot_cmd(void)
8896 {
8897 return register_ftrace_command(&ftrace_snapshot_cmd);
8898 }
8899 #else
register_snapshot_cmd(void)8900 static inline __init int register_snapshot_cmd(void) { return 0; }
8901 #endif /* defined(CONFIG_TRACER_SNAPSHOT) && defined(CONFIG_DYNAMIC_FTRACE) */
8902
tracing_get_dentry(struct trace_array * tr)8903 static struct dentry *tracing_get_dentry(struct trace_array *tr)
8904 {
8905 if (WARN_ON(!tr->dir))
8906 return ERR_PTR(-ENODEV);
8907
8908 /* Top directory uses NULL as the parent */
8909 if (tr->flags & TRACE_ARRAY_FL_GLOBAL)
8910 return NULL;
8911
8912 /* All sub buffers have a descriptor */
8913 return tr->dir;
8914 }
8915
tracing_dentry_percpu(struct trace_array * tr,int cpu)8916 static struct dentry *tracing_dentry_percpu(struct trace_array *tr, int cpu)
8917 {
8918 struct dentry *d_tracer;
8919
8920 if (tr->percpu_dir)
8921 return tr->percpu_dir;
8922
8923 d_tracer = tracing_get_dentry(tr);
8924 if (IS_ERR(d_tracer))
8925 return NULL;
8926
8927 tr->percpu_dir = tracefs_create_dir("per_cpu", d_tracer);
8928
8929 MEM_FAIL(!tr->percpu_dir,
8930 "Could not create tracefs directory 'per_cpu/%d'\n", cpu);
8931
8932 return tr->percpu_dir;
8933 }
8934
8935 static struct dentry *
trace_create_cpu_file(const char * name,umode_t mode,struct dentry * parent,void * data,long cpu,const struct file_operations * fops)8936 trace_create_cpu_file(const char *name, umode_t mode, struct dentry *parent,
8937 void *data, long cpu, const struct file_operations *fops)
8938 {
8939 struct dentry *ret = trace_create_file(name, mode, parent, data, fops);
8940
8941 if (ret) /* See tracing_get_cpu() */
8942 d_inode(ret)->i_cdev = (void *)(cpu + 1);
8943 return ret;
8944 }
8945
8946 static void
tracing_init_tracefs_percpu(struct trace_array * tr,long cpu)8947 tracing_init_tracefs_percpu(struct trace_array *tr, long cpu)
8948 {
8949 struct dentry *d_percpu = tracing_dentry_percpu(tr, cpu);
8950 struct dentry *d_cpu;
8951 char cpu_dir[30]; /* 30 characters should be more than enough */
8952
8953 if (!d_percpu)
8954 return;
8955
8956 snprintf(cpu_dir, 30, "cpu%ld", cpu);
8957 d_cpu = tracefs_create_dir(cpu_dir, d_percpu);
8958 if (!d_cpu) {
8959 pr_warn("Could not create tracefs '%s' entry\n", cpu_dir);
8960 return;
8961 }
8962
8963 /* per cpu trace_pipe */
8964 trace_create_cpu_file("trace_pipe", TRACE_MODE_READ, d_cpu,
8965 tr, cpu, &tracing_pipe_fops);
8966
8967 /* per cpu trace */
8968 trace_create_cpu_file("trace", TRACE_MODE_WRITE, d_cpu,
8969 tr, cpu, &tracing_fops);
8970
8971 trace_create_cpu_file("trace_pipe_raw", TRACE_MODE_READ, d_cpu,
8972 tr, cpu, &tracing_buffers_fops);
8973
8974 trace_create_cpu_file("stats", TRACE_MODE_READ, d_cpu,
8975 tr, cpu, &tracing_stats_fops);
8976
8977 trace_create_cpu_file("buffer_size_kb", TRACE_MODE_READ, d_cpu,
8978 tr, cpu, &tracing_entries_fops);
8979
8980 #ifdef CONFIG_TRACER_SNAPSHOT
8981 trace_create_cpu_file("snapshot", TRACE_MODE_WRITE, d_cpu,
8982 tr, cpu, &snapshot_fops);
8983
8984 trace_create_cpu_file("snapshot_raw", TRACE_MODE_READ, d_cpu,
8985 tr, cpu, &snapshot_raw_fops);
8986 #endif
8987 }
8988
8989 #ifdef CONFIG_FTRACE_SELFTEST
8990 /* Let selftest have access to static functions in this file */
8991 #include "trace_selftest.c"
8992 #endif
8993
8994 static ssize_t
trace_options_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)8995 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
8996 loff_t *ppos)
8997 {
8998 struct trace_option_dentry *topt = filp->private_data;
8999 char *buf;
9000
9001 if (topt->flags->val & topt->opt->bit)
9002 buf = "1\n";
9003 else
9004 buf = "0\n";
9005
9006 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
9007 }
9008
9009 static ssize_t
trace_options_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)9010 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
9011 loff_t *ppos)
9012 {
9013 struct trace_option_dentry *topt = filp->private_data;
9014 unsigned long val;
9015 int ret;
9016
9017 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9018 if (ret)
9019 return ret;
9020
9021 if (val != 0 && val != 1)
9022 return -EINVAL;
9023
9024 if (!!(topt->flags->val & topt->opt->bit) != val) {
9025 mutex_lock(&trace_types_lock);
9026 ret = __set_tracer_option(topt->tr, topt->flags,
9027 topt->opt, !val);
9028 mutex_unlock(&trace_types_lock);
9029 if (ret)
9030 return ret;
9031 }
9032
9033 *ppos += cnt;
9034
9035 return cnt;
9036 }
9037
tracing_open_options(struct inode * inode,struct file * filp)9038 static int tracing_open_options(struct inode *inode, struct file *filp)
9039 {
9040 struct trace_option_dentry *topt = inode->i_private;
9041 int ret;
9042
9043 ret = tracing_check_open_get_tr(topt->tr);
9044 if (ret)
9045 return ret;
9046
9047 filp->private_data = inode->i_private;
9048 return 0;
9049 }
9050
tracing_release_options(struct inode * inode,struct file * file)9051 static int tracing_release_options(struct inode *inode, struct file *file)
9052 {
9053 struct trace_option_dentry *topt = file->private_data;
9054
9055 trace_array_put(topt->tr);
9056 return 0;
9057 }
9058
9059 static const struct file_operations trace_options_fops = {
9060 .open = tracing_open_options,
9061 .read = trace_options_read,
9062 .write = trace_options_write,
9063 .llseek = generic_file_llseek,
9064 .release = tracing_release_options,
9065 };
9066
9067 /*
9068 * In order to pass in both the trace_array descriptor as well as the index
9069 * to the flag that the trace option file represents, the trace_array
9070 * has a character array of trace_flags_index[], which holds the index
9071 * of the bit for the flag it represents. index[0] == 0, index[1] == 1, etc.
9072 * The address of this character array is passed to the flag option file
9073 * read/write callbacks.
9074 *
9075 * In order to extract both the index and the trace_array descriptor,
9076 * get_tr_index() uses the following algorithm.
9077 *
9078 * idx = *ptr;
9079 *
9080 * As the pointer itself contains the address of the index (remember
9081 * index[1] == 1).
9082 *
9083 * Then to get the trace_array descriptor, by subtracting that index
9084 * from the ptr, we get to the start of the index itself.
9085 *
9086 * ptr - idx == &index[0]
9087 *
9088 * Then a simple container_of() from that pointer gets us to the
9089 * trace_array descriptor.
9090 */
get_tr_index(void * data,struct trace_array ** ptr,unsigned int * pindex)9091 static void get_tr_index(void *data, struct trace_array **ptr,
9092 unsigned int *pindex)
9093 {
9094 *pindex = *(unsigned char *)data;
9095
9096 *ptr = container_of(data - *pindex, struct trace_array,
9097 trace_flags_index);
9098 }
9099
9100 static ssize_t
trace_options_core_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)9101 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
9102 loff_t *ppos)
9103 {
9104 void *tr_index = filp->private_data;
9105 struct trace_array *tr;
9106 unsigned int index;
9107 char *buf;
9108
9109 get_tr_index(tr_index, &tr, &index);
9110
9111 if (tr->trace_flags & (1 << index))
9112 buf = "1\n";
9113 else
9114 buf = "0\n";
9115
9116 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
9117 }
9118
9119 static ssize_t
trace_options_core_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)9120 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
9121 loff_t *ppos)
9122 {
9123 void *tr_index = filp->private_data;
9124 struct trace_array *tr;
9125 unsigned int index;
9126 unsigned long val;
9127 int ret;
9128
9129 get_tr_index(tr_index, &tr, &index);
9130
9131 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9132 if (ret)
9133 return ret;
9134
9135 if (val != 0 && val != 1)
9136 return -EINVAL;
9137
9138 mutex_lock(&event_mutex);
9139 mutex_lock(&trace_types_lock);
9140 ret = set_tracer_flag(tr, 1 << index, val);
9141 mutex_unlock(&trace_types_lock);
9142 mutex_unlock(&event_mutex);
9143
9144 if (ret < 0)
9145 return ret;
9146
9147 *ppos += cnt;
9148
9149 return cnt;
9150 }
9151
9152 static const struct file_operations trace_options_core_fops = {
9153 .open = tracing_open_generic,
9154 .read = trace_options_core_read,
9155 .write = trace_options_core_write,
9156 .llseek = generic_file_llseek,
9157 };
9158
trace_create_file(const char * name,umode_t mode,struct dentry * parent,void * data,const struct file_operations * fops)9159 struct dentry *trace_create_file(const char *name,
9160 umode_t mode,
9161 struct dentry *parent,
9162 void *data,
9163 const struct file_operations *fops)
9164 {
9165 struct dentry *ret;
9166
9167 ret = tracefs_create_file(name, mode, parent, data, fops);
9168 if (!ret)
9169 pr_warn("Could not create tracefs '%s' entry\n", name);
9170
9171 return ret;
9172 }
9173
9174
trace_options_init_dentry(struct trace_array * tr)9175 static struct dentry *trace_options_init_dentry(struct trace_array *tr)
9176 {
9177 struct dentry *d_tracer;
9178
9179 if (tr->options)
9180 return tr->options;
9181
9182 d_tracer = tracing_get_dentry(tr);
9183 if (IS_ERR(d_tracer))
9184 return NULL;
9185
9186 tr->options = tracefs_create_dir("options", d_tracer);
9187 if (!tr->options) {
9188 pr_warn("Could not create tracefs directory 'options'\n");
9189 return NULL;
9190 }
9191
9192 return tr->options;
9193 }
9194
9195 static void
create_trace_option_file(struct trace_array * tr,struct trace_option_dentry * topt,struct tracer_flags * flags,struct tracer_opt * opt)9196 create_trace_option_file(struct trace_array *tr,
9197 struct trace_option_dentry *topt,
9198 struct tracer_flags *flags,
9199 struct tracer_opt *opt)
9200 {
9201 struct dentry *t_options;
9202
9203 t_options = trace_options_init_dentry(tr);
9204 if (!t_options)
9205 return;
9206
9207 topt->flags = flags;
9208 topt->opt = opt;
9209 topt->tr = tr;
9210
9211 topt->entry = trace_create_file(opt->name, TRACE_MODE_WRITE,
9212 t_options, topt, &trace_options_fops);
9213
9214 }
9215
9216 static void
create_trace_option_files(struct trace_array * tr,struct tracer * tracer)9217 create_trace_option_files(struct trace_array *tr, struct tracer *tracer)
9218 {
9219 struct trace_option_dentry *topts;
9220 struct trace_options *tr_topts;
9221 struct tracer_flags *flags;
9222 struct tracer_opt *opts;
9223 int cnt;
9224 int i;
9225
9226 if (!tracer)
9227 return;
9228
9229 flags = tracer->flags;
9230
9231 if (!flags || !flags->opts)
9232 return;
9233
9234 /*
9235 * If this is an instance, only create flags for tracers
9236 * the instance may have.
9237 */
9238 if (!trace_ok_for_array(tracer, tr))
9239 return;
9240
9241 for (i = 0; i < tr->nr_topts; i++) {
9242 /* Make sure there's no duplicate flags. */
9243 if (WARN_ON_ONCE(tr->topts[i].tracer->flags == tracer->flags))
9244 return;
9245 }
9246
9247 opts = flags->opts;
9248
9249 for (cnt = 0; opts[cnt].name; cnt++)
9250 ;
9251
9252 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
9253 if (!topts)
9254 return;
9255
9256 tr_topts = krealloc(tr->topts, sizeof(*tr->topts) * (tr->nr_topts + 1),
9257 GFP_KERNEL);
9258 if (!tr_topts) {
9259 kfree(topts);
9260 return;
9261 }
9262
9263 tr->topts = tr_topts;
9264 tr->topts[tr->nr_topts].tracer = tracer;
9265 tr->topts[tr->nr_topts].topts = topts;
9266 tr->nr_topts++;
9267
9268 for (cnt = 0; opts[cnt].name; cnt++) {
9269 create_trace_option_file(tr, &topts[cnt], flags,
9270 &opts[cnt]);
9271 MEM_FAIL(topts[cnt].entry == NULL,
9272 "Failed to create trace option: %s",
9273 opts[cnt].name);
9274 }
9275 }
9276
9277 static struct dentry *
create_trace_option_core_file(struct trace_array * tr,const char * option,long index)9278 create_trace_option_core_file(struct trace_array *tr,
9279 const char *option, long index)
9280 {
9281 struct dentry *t_options;
9282
9283 t_options = trace_options_init_dentry(tr);
9284 if (!t_options)
9285 return NULL;
9286
9287 return trace_create_file(option, TRACE_MODE_WRITE, t_options,
9288 (void *)&tr->trace_flags_index[index],
9289 &trace_options_core_fops);
9290 }
9291
create_trace_options_dir(struct trace_array * tr)9292 static void create_trace_options_dir(struct trace_array *tr)
9293 {
9294 struct dentry *t_options;
9295 bool top_level = tr == &global_trace;
9296 int i;
9297
9298 t_options = trace_options_init_dentry(tr);
9299 if (!t_options)
9300 return;
9301
9302 for (i = 0; trace_options[i]; i++) {
9303 if (top_level ||
9304 !((1 << i) & TOP_LEVEL_TRACE_FLAGS))
9305 create_trace_option_core_file(tr, trace_options[i], i);
9306 }
9307 }
9308
9309 static ssize_t
rb_simple_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)9310 rb_simple_read(struct file *filp, char __user *ubuf,
9311 size_t cnt, loff_t *ppos)
9312 {
9313 struct trace_array *tr = filp->private_data;
9314 char buf[64];
9315 int r;
9316
9317 r = tracer_tracing_is_on(tr);
9318 r = sprintf(buf, "%d\n", r);
9319
9320 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9321 }
9322
9323 static ssize_t
rb_simple_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)9324 rb_simple_write(struct file *filp, const char __user *ubuf,
9325 size_t cnt, loff_t *ppos)
9326 {
9327 struct trace_array *tr = filp->private_data;
9328 struct trace_buffer *buffer = tr->array_buffer.buffer;
9329 unsigned long val;
9330 int ret;
9331
9332 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9333 if (ret)
9334 return ret;
9335
9336 if (buffer) {
9337 mutex_lock(&trace_types_lock);
9338 if (!!val == tracer_tracing_is_on(tr)) {
9339 val = 0; /* do nothing */
9340 } else if (val) {
9341 tracer_tracing_on(tr);
9342 if (tr->current_trace->start)
9343 tr->current_trace->start(tr);
9344 } else {
9345 tracer_tracing_off(tr);
9346 if (tr->current_trace->stop)
9347 tr->current_trace->stop(tr);
9348 /* Wake up any waiters */
9349 ring_buffer_wake_waiters(buffer, RING_BUFFER_ALL_CPUS);
9350 }
9351 mutex_unlock(&trace_types_lock);
9352 }
9353
9354 (*ppos)++;
9355
9356 return cnt;
9357 }
9358
9359 static const struct file_operations rb_simple_fops = {
9360 .open = tracing_open_generic_tr,
9361 .read = rb_simple_read,
9362 .write = rb_simple_write,
9363 .release = tracing_release_generic_tr,
9364 .llseek = default_llseek,
9365 };
9366
9367 static ssize_t
buffer_percent_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)9368 buffer_percent_read(struct file *filp, char __user *ubuf,
9369 size_t cnt, loff_t *ppos)
9370 {
9371 struct trace_array *tr = filp->private_data;
9372 char buf[64];
9373 int r;
9374
9375 r = tr->buffer_percent;
9376 r = sprintf(buf, "%d\n", r);
9377
9378 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9379 }
9380
9381 static ssize_t
buffer_percent_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)9382 buffer_percent_write(struct file *filp, const char __user *ubuf,
9383 size_t cnt, loff_t *ppos)
9384 {
9385 struct trace_array *tr = filp->private_data;
9386 unsigned long val;
9387 int ret;
9388
9389 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9390 if (ret)
9391 return ret;
9392
9393 if (val > 100)
9394 return -EINVAL;
9395
9396 tr->buffer_percent = val;
9397
9398 (*ppos)++;
9399
9400 return cnt;
9401 }
9402
9403 static const struct file_operations buffer_percent_fops = {
9404 .open = tracing_open_generic_tr,
9405 .read = buffer_percent_read,
9406 .write = buffer_percent_write,
9407 .release = tracing_release_generic_tr,
9408 .llseek = default_llseek,
9409 };
9410
9411 static ssize_t
buffer_subbuf_size_read(struct file * filp,char __user * ubuf,size_t cnt,loff_t * ppos)9412 buffer_subbuf_size_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
9413 {
9414 struct trace_array *tr = filp->private_data;
9415 size_t size;
9416 char buf[64];
9417 int order;
9418 int r;
9419
9420 order = ring_buffer_subbuf_order_get(tr->array_buffer.buffer);
9421 size = (PAGE_SIZE << order) / 1024;
9422
9423 r = sprintf(buf, "%zd\n", size);
9424
9425 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
9426 }
9427
9428 static ssize_t
buffer_subbuf_size_write(struct file * filp,const char __user * ubuf,size_t cnt,loff_t * ppos)9429 buffer_subbuf_size_write(struct file *filp, const char __user *ubuf,
9430 size_t cnt, loff_t *ppos)
9431 {
9432 struct trace_array *tr = filp->private_data;
9433 unsigned long val;
9434 int old_order;
9435 int order;
9436 int pages;
9437 int ret;
9438
9439 ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
9440 if (ret)
9441 return ret;
9442
9443 val *= 1024; /* value passed in is in KB */
9444
9445 pages = DIV_ROUND_UP(val, PAGE_SIZE);
9446 order = fls(pages - 1);
9447
9448 /* limit between 1 and 128 system pages */
9449 if (order < 0 || order > 7)
9450 return -EINVAL;
9451
9452 /* Do not allow tracing while changing the order of the ring buffer */
9453 tracing_stop_tr(tr);
9454
9455 old_order = ring_buffer_subbuf_order_get(tr->array_buffer.buffer);
9456 if (old_order == order)
9457 goto out;
9458
9459 ret = ring_buffer_subbuf_order_set(tr->array_buffer.buffer, order);
9460 if (ret)
9461 goto out;
9462
9463 #ifdef CONFIG_TRACER_MAX_TRACE
9464
9465 if (!tr->allocated_snapshot)
9466 goto out_max;
9467
9468 ret = ring_buffer_subbuf_order_set(tr->max_buffer.buffer, order);
9469 if (ret) {
9470 /* Put back the old order */
9471 cnt = ring_buffer_subbuf_order_set(tr->array_buffer.buffer, old_order);
9472 if (WARN_ON_ONCE(cnt)) {
9473 /*
9474 * AARGH! We are left with different orders!
9475 * The max buffer is our "snapshot" buffer.
9476 * When a tracer needs a snapshot (one of the
9477 * latency tracers), it swaps the max buffer
9478 * with the saved snap shot. We succeeded to
9479 * update the order of the main buffer, but failed to
9480 * update the order of the max buffer. But when we tried
9481 * to reset the main buffer to the original size, we
9482 * failed there too. This is very unlikely to
9483 * happen, but if it does, warn and kill all
9484 * tracing.
9485 */
9486 tracing_disabled = 1;
9487 }
9488 goto out;
9489 }
9490 out_max:
9491 #endif
9492 (*ppos)++;
9493 out:
9494 if (ret)
9495 cnt = ret;
9496 tracing_start_tr(tr);
9497 return cnt;
9498 }
9499
9500 static const struct file_operations buffer_subbuf_size_fops = {
9501 .open = tracing_open_generic_tr,
9502 .read = buffer_subbuf_size_read,
9503 .write = buffer_subbuf_size_write,
9504 .release = tracing_release_generic_tr,
9505 .llseek = default_llseek,
9506 };
9507
9508 static struct dentry *trace_instance_dir;
9509
9510 static void
9511 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer);
9512
9513 static int
allocate_trace_buffer(struct trace_array * tr,struct array_buffer * buf,int size)9514 allocate_trace_buffer(struct trace_array *tr, struct array_buffer *buf, int size)
9515 {
9516 enum ring_buffer_flags rb_flags;
9517
9518 rb_flags = tr->trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
9519
9520 buf->tr = tr;
9521
9522 buf->buffer = ring_buffer_alloc(size, rb_flags);
9523 if (!buf->buffer)
9524 return -ENOMEM;
9525
9526 buf->data = alloc_percpu(struct trace_array_cpu);
9527 if (!buf->data) {
9528 ring_buffer_free(buf->buffer);
9529 buf->buffer = NULL;
9530 return -ENOMEM;
9531 }
9532
9533 /* Allocate the first page for all buffers */
9534 set_buffer_entries(&tr->array_buffer,
9535 ring_buffer_size(tr->array_buffer.buffer, 0));
9536
9537 return 0;
9538 }
9539
free_trace_buffer(struct array_buffer * buf)9540 static void free_trace_buffer(struct array_buffer *buf)
9541 {
9542 if (buf->buffer) {
9543 ring_buffer_free(buf->buffer);
9544 buf->buffer = NULL;
9545 free_percpu(buf->data);
9546 buf->data = NULL;
9547 }
9548 }
9549
allocate_trace_buffers(struct trace_array * tr,int size)9550 static int allocate_trace_buffers(struct trace_array *tr, int size)
9551 {
9552 int ret;
9553
9554 ret = allocate_trace_buffer(tr, &tr->array_buffer, size);
9555 if (ret)
9556 return ret;
9557
9558 #ifdef CONFIG_TRACER_MAX_TRACE
9559 ret = allocate_trace_buffer(tr, &tr->max_buffer,
9560 allocate_snapshot ? size : 1);
9561 if (MEM_FAIL(ret, "Failed to allocate trace buffer\n")) {
9562 free_trace_buffer(&tr->array_buffer);
9563 return -ENOMEM;
9564 }
9565 tr->allocated_snapshot = allocate_snapshot;
9566
9567 allocate_snapshot = false;
9568 #endif
9569
9570 return 0;
9571 }
9572
free_trace_buffers(struct trace_array * tr)9573 static void free_trace_buffers(struct trace_array *tr)
9574 {
9575 if (!tr)
9576 return;
9577
9578 free_trace_buffer(&tr->array_buffer);
9579
9580 #ifdef CONFIG_TRACER_MAX_TRACE
9581 free_trace_buffer(&tr->max_buffer);
9582 #endif
9583 }
9584
init_trace_flags_index(struct trace_array * tr)9585 static void init_trace_flags_index(struct trace_array *tr)
9586 {
9587 int i;
9588
9589 /* Used by the trace options files */
9590 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++)
9591 tr->trace_flags_index[i] = i;
9592 }
9593
__update_tracer_options(struct trace_array * tr)9594 static void __update_tracer_options(struct trace_array *tr)
9595 {
9596 struct tracer *t;
9597
9598 for (t = trace_types; t; t = t->next)
9599 add_tracer_options(tr, t);
9600 }
9601
update_tracer_options(struct trace_array * tr)9602 static void update_tracer_options(struct trace_array *tr)
9603 {
9604 mutex_lock(&trace_types_lock);
9605 tracer_options_updated = true;
9606 __update_tracer_options(tr);
9607 mutex_unlock(&trace_types_lock);
9608 }
9609
9610 /* Must have trace_types_lock held */
trace_array_find(const char * instance)9611 struct trace_array *trace_array_find(const char *instance)
9612 {
9613 struct trace_array *tr, *found = NULL;
9614
9615 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9616 if (tr->name && strcmp(tr->name, instance) == 0) {
9617 found = tr;
9618 break;
9619 }
9620 }
9621
9622 return found;
9623 }
9624
trace_array_find_get(const char * instance)9625 struct trace_array *trace_array_find_get(const char *instance)
9626 {
9627 struct trace_array *tr;
9628
9629 mutex_lock(&trace_types_lock);
9630 tr = trace_array_find(instance);
9631 if (tr)
9632 tr->ref++;
9633 mutex_unlock(&trace_types_lock);
9634
9635 return tr;
9636 }
9637
trace_array_create_dir(struct trace_array * tr)9638 static int trace_array_create_dir(struct trace_array *tr)
9639 {
9640 int ret;
9641
9642 tr->dir = tracefs_create_dir(tr->name, trace_instance_dir);
9643 if (!tr->dir)
9644 return -EINVAL;
9645
9646 ret = event_trace_add_tracer(tr->dir, tr);
9647 if (ret) {
9648 tracefs_remove(tr->dir);
9649 return ret;
9650 }
9651
9652 init_tracer_tracefs(tr, tr->dir);
9653 __update_tracer_options(tr);
9654
9655 return ret;
9656 }
9657
9658 static struct trace_array *
trace_array_create_systems(const char * name,const char * systems)9659 trace_array_create_systems(const char *name, const char *systems)
9660 {
9661 struct trace_array *tr;
9662 int ret;
9663
9664 ret = -ENOMEM;
9665 tr = kzalloc(sizeof(*tr), GFP_KERNEL);
9666 if (!tr)
9667 return ERR_PTR(ret);
9668
9669 tr->name = kstrdup(name, GFP_KERNEL);
9670 if (!tr->name)
9671 goto out_free_tr;
9672
9673 if (!alloc_cpumask_var(&tr->tracing_cpumask, GFP_KERNEL))
9674 goto out_free_tr;
9675
9676 if (!zalloc_cpumask_var(&tr->pipe_cpumask, GFP_KERNEL))
9677 goto out_free_tr;
9678
9679 if (systems) {
9680 tr->system_names = kstrdup_const(systems, GFP_KERNEL);
9681 if (!tr->system_names)
9682 goto out_free_tr;
9683 }
9684
9685 tr->trace_flags = global_trace.trace_flags & ~ZEROED_TRACE_FLAGS;
9686
9687 cpumask_copy(tr->tracing_cpumask, cpu_all_mask);
9688
9689 raw_spin_lock_init(&tr->start_lock);
9690
9691 tr->max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
9692
9693 tr->current_trace = &nop_trace;
9694
9695 INIT_LIST_HEAD(&tr->systems);
9696 INIT_LIST_HEAD(&tr->events);
9697 INIT_LIST_HEAD(&tr->hist_vars);
9698 INIT_LIST_HEAD(&tr->err_log);
9699
9700 if (allocate_trace_buffers(tr, trace_buf_size) < 0)
9701 goto out_free_tr;
9702
9703 /* The ring buffer is defaultly expanded */
9704 trace_set_ring_buffer_expanded(tr);
9705
9706 if (ftrace_allocate_ftrace_ops(tr) < 0)
9707 goto out_free_tr;
9708
9709 ftrace_init_trace_array(tr);
9710
9711 init_trace_flags_index(tr);
9712
9713 if (trace_instance_dir) {
9714 ret = trace_array_create_dir(tr);
9715 if (ret)
9716 goto out_free_tr;
9717 } else
9718 __trace_early_add_events(tr);
9719
9720 list_add(&tr->list, &ftrace_trace_arrays);
9721
9722 tr->ref++;
9723
9724 return tr;
9725
9726 out_free_tr:
9727 ftrace_free_ftrace_ops(tr);
9728 free_trace_buffers(tr);
9729 free_cpumask_var(tr->pipe_cpumask);
9730 free_cpumask_var(tr->tracing_cpumask);
9731 kfree_const(tr->system_names);
9732 kfree(tr->name);
9733 kfree(tr);
9734
9735 return ERR_PTR(ret);
9736 }
9737
trace_array_create(const char * name)9738 static struct trace_array *trace_array_create(const char *name)
9739 {
9740 return trace_array_create_systems(name, NULL);
9741 }
9742
instance_mkdir(const char * name)9743 static int instance_mkdir(const char *name)
9744 {
9745 struct trace_array *tr;
9746 int ret;
9747
9748 mutex_lock(&event_mutex);
9749 mutex_lock(&trace_types_lock);
9750
9751 ret = -EEXIST;
9752 if (trace_array_find(name))
9753 goto out_unlock;
9754
9755 tr = trace_array_create(name);
9756
9757 ret = PTR_ERR_OR_ZERO(tr);
9758
9759 out_unlock:
9760 mutex_unlock(&trace_types_lock);
9761 mutex_unlock(&event_mutex);
9762 return ret;
9763 }
9764
9765 /**
9766 * trace_array_get_by_name - Create/Lookup a trace array, given its name.
9767 * @name: The name of the trace array to be looked up/created.
9768 * @systems: A list of systems to create event directories for (NULL for all)
9769 *
9770 * Returns pointer to trace array with given name.
9771 * NULL, if it cannot be created.
9772 *
9773 * NOTE: This function increments the reference counter associated with the
9774 * trace array returned. This makes sure it cannot be freed while in use.
9775 * Use trace_array_put() once the trace array is no longer needed.
9776 * If the trace_array is to be freed, trace_array_destroy() needs to
9777 * be called after the trace_array_put(), or simply let user space delete
9778 * it from the tracefs instances directory. But until the
9779 * trace_array_put() is called, user space can not delete it.
9780 *
9781 */
trace_array_get_by_name(const char * name,const char * systems)9782 struct trace_array *trace_array_get_by_name(const char *name, const char *systems)
9783 {
9784 struct trace_array *tr;
9785
9786 mutex_lock(&event_mutex);
9787 mutex_lock(&trace_types_lock);
9788
9789 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9790 if (tr->name && strcmp(tr->name, name) == 0)
9791 goto out_unlock;
9792 }
9793
9794 tr = trace_array_create_systems(name, systems);
9795
9796 if (IS_ERR(tr))
9797 tr = NULL;
9798 out_unlock:
9799 if (tr)
9800 tr->ref++;
9801
9802 mutex_unlock(&trace_types_lock);
9803 mutex_unlock(&event_mutex);
9804 return tr;
9805 }
9806 EXPORT_SYMBOL_GPL(trace_array_get_by_name);
9807
__remove_instance(struct trace_array * tr)9808 static int __remove_instance(struct trace_array *tr)
9809 {
9810 int i;
9811
9812 /* Reference counter for a newly created trace array = 1. */
9813 if (tr->ref > 1 || (tr->current_trace && tr->trace_ref))
9814 return -EBUSY;
9815
9816 list_del(&tr->list);
9817
9818 /* Disable all the flags that were enabled coming in */
9819 for (i = 0; i < TRACE_FLAGS_MAX_SIZE; i++) {
9820 if ((1 << i) & ZEROED_TRACE_FLAGS)
9821 set_tracer_flag(tr, 1 << i, 0);
9822 }
9823
9824 tracing_set_nop(tr);
9825 clear_ftrace_function_probes(tr);
9826 event_trace_del_tracer(tr);
9827 ftrace_clear_pids(tr);
9828 ftrace_destroy_function_files(tr);
9829 tracefs_remove(tr->dir);
9830 free_percpu(tr->last_func_repeats);
9831 free_trace_buffers(tr);
9832 clear_tracing_err_log(tr);
9833
9834 for (i = 0; i < tr->nr_topts; i++) {
9835 kfree(tr->topts[i].topts);
9836 }
9837 kfree(tr->topts);
9838
9839 free_cpumask_var(tr->pipe_cpumask);
9840 free_cpumask_var(tr->tracing_cpumask);
9841 kfree_const(tr->system_names);
9842 kfree(tr->name);
9843 kfree(tr);
9844
9845 return 0;
9846 }
9847
trace_array_destroy(struct trace_array * this_tr)9848 int trace_array_destroy(struct trace_array *this_tr)
9849 {
9850 struct trace_array *tr;
9851 int ret;
9852
9853 if (!this_tr)
9854 return -EINVAL;
9855
9856 mutex_lock(&event_mutex);
9857 mutex_lock(&trace_types_lock);
9858
9859 ret = -ENODEV;
9860
9861 /* Making sure trace array exists before destroying it. */
9862 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9863 if (tr == this_tr) {
9864 ret = __remove_instance(tr);
9865 break;
9866 }
9867 }
9868
9869 mutex_unlock(&trace_types_lock);
9870 mutex_unlock(&event_mutex);
9871
9872 return ret;
9873 }
9874 EXPORT_SYMBOL_GPL(trace_array_destroy);
9875
instance_rmdir(const char * name)9876 static int instance_rmdir(const char *name)
9877 {
9878 struct trace_array *tr;
9879 int ret;
9880
9881 mutex_lock(&event_mutex);
9882 mutex_lock(&trace_types_lock);
9883
9884 ret = -ENODEV;
9885 tr = trace_array_find(name);
9886 if (tr)
9887 ret = __remove_instance(tr);
9888
9889 mutex_unlock(&trace_types_lock);
9890 mutex_unlock(&event_mutex);
9891
9892 return ret;
9893 }
9894
create_trace_instances(struct dentry * d_tracer)9895 static __init void create_trace_instances(struct dentry *d_tracer)
9896 {
9897 struct trace_array *tr;
9898
9899 trace_instance_dir = tracefs_create_instance_dir("instances", d_tracer,
9900 instance_mkdir,
9901 instance_rmdir);
9902 if (MEM_FAIL(!trace_instance_dir, "Failed to create instances directory\n"))
9903 return;
9904
9905 mutex_lock(&event_mutex);
9906 mutex_lock(&trace_types_lock);
9907
9908 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
9909 if (!tr->name)
9910 continue;
9911 if (MEM_FAIL(trace_array_create_dir(tr) < 0,
9912 "Failed to create instance directory\n"))
9913 break;
9914 }
9915
9916 mutex_unlock(&trace_types_lock);
9917 mutex_unlock(&event_mutex);
9918 }
9919
9920 static void
init_tracer_tracefs(struct trace_array * tr,struct dentry * d_tracer)9921 init_tracer_tracefs(struct trace_array *tr, struct dentry *d_tracer)
9922 {
9923 int cpu;
9924
9925 trace_create_file("available_tracers", TRACE_MODE_READ, d_tracer,
9926 tr, &show_traces_fops);
9927
9928 trace_create_file("current_tracer", TRACE_MODE_WRITE, d_tracer,
9929 tr, &set_tracer_fops);
9930
9931 trace_create_file("tracing_cpumask", TRACE_MODE_WRITE, d_tracer,
9932 tr, &tracing_cpumask_fops);
9933
9934 trace_create_file("trace_options", TRACE_MODE_WRITE, d_tracer,
9935 tr, &tracing_iter_fops);
9936
9937 trace_create_file("trace", TRACE_MODE_WRITE, d_tracer,
9938 tr, &tracing_fops);
9939
9940 trace_create_file("trace_pipe", TRACE_MODE_READ, d_tracer,
9941 tr, &tracing_pipe_fops);
9942
9943 trace_create_file("buffer_size_kb", TRACE_MODE_WRITE, d_tracer,
9944 tr, &tracing_entries_fops);
9945
9946 trace_create_file("buffer_total_size_kb", TRACE_MODE_READ, d_tracer,
9947 tr, &tracing_total_entries_fops);
9948
9949 trace_create_file("free_buffer", 0200, d_tracer,
9950 tr, &tracing_free_buffer_fops);
9951
9952 trace_create_file("trace_marker", 0220, d_tracer,
9953 tr, &tracing_mark_fops);
9954
9955 tr->trace_marker_file = __find_event_file(tr, "ftrace", "print");
9956
9957 trace_create_file("trace_marker_raw", 0220, d_tracer,
9958 tr, &tracing_mark_raw_fops);
9959
9960 trace_create_file("trace_clock", TRACE_MODE_WRITE, d_tracer, tr,
9961 &trace_clock_fops);
9962
9963 trace_create_file("tracing_on", TRACE_MODE_WRITE, d_tracer,
9964 tr, &rb_simple_fops);
9965
9966 trace_create_file("timestamp_mode", TRACE_MODE_READ, d_tracer, tr,
9967 &trace_time_stamp_mode_fops);
9968
9969 tr->buffer_percent = 50;
9970
9971 trace_create_file("buffer_percent", TRACE_MODE_WRITE, d_tracer,
9972 tr, &buffer_percent_fops);
9973
9974 trace_create_file("buffer_subbuf_size_kb", TRACE_MODE_WRITE, d_tracer,
9975 tr, &buffer_subbuf_size_fops);
9976
9977 create_trace_options_dir(tr);
9978
9979 #ifdef CONFIG_TRACER_MAX_TRACE
9980 trace_create_maxlat_file(tr, d_tracer);
9981 #endif
9982
9983 if (ftrace_create_function_files(tr, d_tracer))
9984 MEM_FAIL(1, "Could not allocate function filter files");
9985
9986 #ifdef CONFIG_TRACER_SNAPSHOT
9987 trace_create_file("snapshot", TRACE_MODE_WRITE, d_tracer,
9988 tr, &snapshot_fops);
9989 #endif
9990
9991 trace_create_file("error_log", TRACE_MODE_WRITE, d_tracer,
9992 tr, &tracing_err_log_fops);
9993
9994 for_each_tracing_cpu(cpu)
9995 tracing_init_tracefs_percpu(tr, cpu);
9996
9997 ftrace_init_tracefs(tr, d_tracer);
9998 }
9999
trace_automount(struct dentry * mntpt,void * ingore)10000 static struct vfsmount *trace_automount(struct dentry *mntpt, void *ingore)
10001 {
10002 struct vfsmount *mnt;
10003 struct file_system_type *type;
10004
10005 /*
10006 * To maintain backward compatibility for tools that mount
10007 * debugfs to get to the tracing facility, tracefs is automatically
10008 * mounted to the debugfs/tracing directory.
10009 */
10010 type = get_fs_type("tracefs");
10011 if (!type)
10012 return NULL;
10013 mnt = vfs_submount(mntpt, type, "tracefs", NULL);
10014 put_filesystem(type);
10015 if (IS_ERR(mnt))
10016 return NULL;
10017 mntget(mnt);
10018
10019 return mnt;
10020 }
10021
10022 /**
10023 * tracing_init_dentry - initialize top level trace array
10024 *
10025 * This is called when creating files or directories in the tracing
10026 * directory. It is called via fs_initcall() by any of the boot up code
10027 * and expects to return the dentry of the top level tracing directory.
10028 */
tracing_init_dentry(void)10029 int tracing_init_dentry(void)
10030 {
10031 struct trace_array *tr = &global_trace;
10032
10033 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10034 pr_warn("Tracing disabled due to lockdown\n");
10035 return -EPERM;
10036 }
10037
10038 /* The top level trace array uses NULL as parent */
10039 if (tr->dir)
10040 return 0;
10041
10042 if (WARN_ON(!tracefs_initialized()))
10043 return -ENODEV;
10044
10045 /*
10046 * As there may still be users that expect the tracing
10047 * files to exist in debugfs/tracing, we must automount
10048 * the tracefs file system there, so older tools still
10049 * work with the newer kernel.
10050 */
10051 tr->dir = debugfs_create_automount("tracing", NULL,
10052 trace_automount, NULL);
10053
10054 return 0;
10055 }
10056
10057 extern struct trace_eval_map *__start_ftrace_eval_maps[];
10058 extern struct trace_eval_map *__stop_ftrace_eval_maps[];
10059
10060 static struct workqueue_struct *eval_map_wq __initdata;
10061 static struct work_struct eval_map_work __initdata;
10062 static struct work_struct tracerfs_init_work __initdata;
10063
eval_map_work_func(struct work_struct * work)10064 static void __init eval_map_work_func(struct work_struct *work)
10065 {
10066 int len;
10067
10068 len = __stop_ftrace_eval_maps - __start_ftrace_eval_maps;
10069 trace_insert_eval_map(NULL, __start_ftrace_eval_maps, len);
10070 }
10071
trace_eval_init(void)10072 static int __init trace_eval_init(void)
10073 {
10074 INIT_WORK(&eval_map_work, eval_map_work_func);
10075
10076 eval_map_wq = alloc_workqueue("eval_map_wq", WQ_UNBOUND, 0);
10077 if (!eval_map_wq) {
10078 pr_err("Unable to allocate eval_map_wq\n");
10079 /* Do work here */
10080 eval_map_work_func(&eval_map_work);
10081 return -ENOMEM;
10082 }
10083
10084 queue_work(eval_map_wq, &eval_map_work);
10085 return 0;
10086 }
10087
10088 subsys_initcall(trace_eval_init);
10089
trace_eval_sync(void)10090 static int __init trace_eval_sync(void)
10091 {
10092 /* Make sure the eval map updates are finished */
10093 if (eval_map_wq)
10094 destroy_workqueue(eval_map_wq);
10095 return 0;
10096 }
10097
10098 late_initcall_sync(trace_eval_sync);
10099
10100
10101 #ifdef CONFIG_MODULES
trace_module_add_evals(struct module * mod)10102 static void trace_module_add_evals(struct module *mod)
10103 {
10104 if (!mod->num_trace_evals)
10105 return;
10106
10107 /*
10108 * Modules with bad taint do not have events created, do
10109 * not bother with enums either.
10110 */
10111 if (trace_module_has_bad_taint(mod))
10112 return;
10113
10114 trace_insert_eval_map(mod, mod->trace_evals, mod->num_trace_evals);
10115 }
10116
10117 #ifdef CONFIG_TRACE_EVAL_MAP_FILE
trace_module_remove_evals(struct module * mod)10118 static void trace_module_remove_evals(struct module *mod)
10119 {
10120 union trace_eval_map_item *map;
10121 union trace_eval_map_item **last = &trace_eval_maps;
10122
10123 if (!mod->num_trace_evals)
10124 return;
10125
10126 mutex_lock(&trace_eval_mutex);
10127
10128 map = trace_eval_maps;
10129
10130 while (map) {
10131 if (map->head.mod == mod)
10132 break;
10133 map = trace_eval_jmp_to_tail(map);
10134 last = &map->tail.next;
10135 map = map->tail.next;
10136 }
10137 if (!map)
10138 goto out;
10139
10140 *last = trace_eval_jmp_to_tail(map)->tail.next;
10141 kfree(map);
10142 out:
10143 mutex_unlock(&trace_eval_mutex);
10144 }
10145 #else
trace_module_remove_evals(struct module * mod)10146 static inline void trace_module_remove_evals(struct module *mod) { }
10147 #endif /* CONFIG_TRACE_EVAL_MAP_FILE */
10148
trace_module_notify(struct notifier_block * self,unsigned long val,void * data)10149 static int trace_module_notify(struct notifier_block *self,
10150 unsigned long val, void *data)
10151 {
10152 struct module *mod = data;
10153
10154 switch (val) {
10155 case MODULE_STATE_COMING:
10156 trace_module_add_evals(mod);
10157 break;
10158 case MODULE_STATE_GOING:
10159 trace_module_remove_evals(mod);
10160 break;
10161 }
10162
10163 return NOTIFY_OK;
10164 }
10165
10166 static struct notifier_block trace_module_nb = {
10167 .notifier_call = trace_module_notify,
10168 .priority = 0,
10169 };
10170 #endif /* CONFIG_MODULES */
10171
tracer_init_tracefs_work_func(struct work_struct * work)10172 static __init void tracer_init_tracefs_work_func(struct work_struct *work)
10173 {
10174
10175 event_trace_init();
10176
10177 init_tracer_tracefs(&global_trace, NULL);
10178 ftrace_init_tracefs_toplevel(&global_trace, NULL);
10179
10180 trace_create_file("tracing_thresh", TRACE_MODE_WRITE, NULL,
10181 &global_trace, &tracing_thresh_fops);
10182
10183 trace_create_file("README", TRACE_MODE_READ, NULL,
10184 NULL, &tracing_readme_fops);
10185
10186 trace_create_file("saved_cmdlines", TRACE_MODE_READ, NULL,
10187 NULL, &tracing_saved_cmdlines_fops);
10188
10189 trace_create_file("saved_cmdlines_size", TRACE_MODE_WRITE, NULL,
10190 NULL, &tracing_saved_cmdlines_size_fops);
10191
10192 trace_create_file("saved_tgids", TRACE_MODE_READ, NULL,
10193 NULL, &tracing_saved_tgids_fops);
10194
10195 trace_create_eval_file(NULL);
10196
10197 #ifdef CONFIG_MODULES
10198 register_module_notifier(&trace_module_nb);
10199 #endif
10200
10201 #ifdef CONFIG_DYNAMIC_FTRACE
10202 trace_create_file("dyn_ftrace_total_info", TRACE_MODE_READ, NULL,
10203 NULL, &tracing_dyn_info_fops);
10204 #endif
10205
10206 create_trace_instances(NULL);
10207
10208 update_tracer_options(&global_trace);
10209 }
10210
tracer_init_tracefs(void)10211 static __init int tracer_init_tracefs(void)
10212 {
10213 int ret;
10214
10215 trace_access_lock_init();
10216
10217 ret = tracing_init_dentry();
10218 if (ret)
10219 return 0;
10220
10221 if (eval_map_wq) {
10222 INIT_WORK(&tracerfs_init_work, tracer_init_tracefs_work_func);
10223 queue_work(eval_map_wq, &tracerfs_init_work);
10224 } else {
10225 tracer_init_tracefs_work_func(NULL);
10226 }
10227
10228 rv_init_interface();
10229
10230 return 0;
10231 }
10232
10233 fs_initcall(tracer_init_tracefs);
10234
10235 static int trace_die_panic_handler(struct notifier_block *self,
10236 unsigned long ev, void *unused);
10237
10238 static struct notifier_block trace_panic_notifier = {
10239 .notifier_call = trace_die_panic_handler,
10240 .priority = INT_MAX - 1,
10241 };
10242
10243 static struct notifier_block trace_die_notifier = {
10244 .notifier_call = trace_die_panic_handler,
10245 .priority = INT_MAX - 1,
10246 };
10247
10248 /*
10249 * The idea is to execute the following die/panic callback early, in order
10250 * to avoid showing irrelevant information in the trace (like other panic
10251 * notifier functions); we are the 2nd to run, after hung_task/rcu_stall
10252 * warnings get disabled (to prevent potential log flooding).
10253 */
trace_die_panic_handler(struct notifier_block * self,unsigned long ev,void * unused)10254 static int trace_die_panic_handler(struct notifier_block *self,
10255 unsigned long ev, void *unused)
10256 {
10257 if (!ftrace_dump_on_oops)
10258 return NOTIFY_DONE;
10259
10260 /* The die notifier requires DIE_OOPS to trigger */
10261 if (self == &trace_die_notifier && ev != DIE_OOPS)
10262 return NOTIFY_DONE;
10263
10264 ftrace_dump(ftrace_dump_on_oops);
10265
10266 return NOTIFY_DONE;
10267 }
10268
10269 /*
10270 * printk is set to max of 1024, we really don't need it that big.
10271 * Nothing should be printing 1000 characters anyway.
10272 */
10273 #define TRACE_MAX_PRINT 1000
10274
10275 /*
10276 * Define here KERN_TRACE so that we have one place to modify
10277 * it if we decide to change what log level the ftrace dump
10278 * should be at.
10279 */
10280 #define KERN_TRACE KERN_EMERG
10281
10282 void
trace_printk_seq(struct trace_seq * s)10283 trace_printk_seq(struct trace_seq *s)
10284 {
10285 /* Probably should print a warning here. */
10286 if (s->seq.len >= TRACE_MAX_PRINT)
10287 s->seq.len = TRACE_MAX_PRINT;
10288
10289 /*
10290 * More paranoid code. Although the buffer size is set to
10291 * PAGE_SIZE, and TRACE_MAX_PRINT is 1000, this is just
10292 * an extra layer of protection.
10293 */
10294 if (WARN_ON_ONCE(s->seq.len >= s->seq.size))
10295 s->seq.len = s->seq.size - 1;
10296
10297 /* should be zero ended, but we are paranoid. */
10298 s->buffer[s->seq.len] = 0;
10299
10300 printk(KERN_TRACE "%s", s->buffer);
10301
10302 trace_seq_init(s);
10303 }
10304
trace_init_global_iter(struct trace_iterator * iter)10305 void trace_init_global_iter(struct trace_iterator *iter)
10306 {
10307 iter->tr = &global_trace;
10308 iter->trace = iter->tr->current_trace;
10309 iter->cpu_file = RING_BUFFER_ALL_CPUS;
10310 iter->array_buffer = &global_trace.array_buffer;
10311
10312 if (iter->trace && iter->trace->open)
10313 iter->trace->open(iter);
10314
10315 /* Annotate start of buffers if we had overruns */
10316 if (ring_buffer_overruns(iter->array_buffer->buffer))
10317 iter->iter_flags |= TRACE_FILE_ANNOTATE;
10318
10319 /* Output in nanoseconds only if we are using a clock in nanoseconds. */
10320 if (trace_clocks[iter->tr->clock_id].in_ns)
10321 iter->iter_flags |= TRACE_FILE_TIME_IN_NS;
10322
10323 /* Can not use kmalloc for iter.temp and iter.fmt */
10324 iter->temp = static_temp_buf;
10325 iter->temp_size = STATIC_TEMP_BUF_SIZE;
10326 iter->fmt = static_fmt_buf;
10327 iter->fmt_size = STATIC_FMT_BUF_SIZE;
10328 }
10329
ftrace_dump(enum ftrace_dump_mode oops_dump_mode)10330 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
10331 {
10332 /* use static because iter can be a bit big for the stack */
10333 static struct trace_iterator iter;
10334 static atomic_t dump_running;
10335 struct trace_array *tr = &global_trace;
10336 unsigned int old_userobj;
10337 unsigned long flags;
10338 int cnt = 0, cpu;
10339
10340 /* Only allow one dump user at a time. */
10341 if (atomic_inc_return(&dump_running) != 1) {
10342 atomic_dec(&dump_running);
10343 return;
10344 }
10345
10346 /*
10347 * Always turn off tracing when we dump.
10348 * We don't need to show trace output of what happens
10349 * between multiple crashes.
10350 *
10351 * If the user does a sysrq-z, then they can re-enable
10352 * tracing with echo 1 > tracing_on.
10353 */
10354 tracing_off();
10355
10356 local_irq_save(flags);
10357
10358 /* Simulate the iterator */
10359 trace_init_global_iter(&iter);
10360
10361 for_each_tracing_cpu(cpu) {
10362 atomic_inc(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10363 }
10364
10365 old_userobj = tr->trace_flags & TRACE_ITER_SYM_USEROBJ;
10366
10367 /* don't look at user memory in panic mode */
10368 tr->trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
10369
10370 switch (oops_dump_mode) {
10371 case DUMP_ALL:
10372 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10373 break;
10374 case DUMP_ORIG:
10375 iter.cpu_file = raw_smp_processor_id();
10376 break;
10377 case DUMP_NONE:
10378 goto out_enable;
10379 default:
10380 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
10381 iter.cpu_file = RING_BUFFER_ALL_CPUS;
10382 }
10383
10384 printk(KERN_TRACE "Dumping ftrace buffer:\n");
10385
10386 /* Did function tracer already get disabled? */
10387 if (ftrace_is_dead()) {
10388 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
10389 printk("# MAY BE MISSING FUNCTION EVENTS\n");
10390 }
10391
10392 /*
10393 * We need to stop all tracing on all CPUS to read
10394 * the next buffer. This is a bit expensive, but is
10395 * not done often. We fill all what we can read,
10396 * and then release the locks again.
10397 */
10398
10399 while (!trace_empty(&iter)) {
10400
10401 if (!cnt)
10402 printk(KERN_TRACE "---------------------------------\n");
10403
10404 cnt++;
10405
10406 trace_iterator_reset(&iter);
10407 iter.iter_flags |= TRACE_FILE_LAT_FMT;
10408
10409 if (trace_find_next_entry_inc(&iter) != NULL) {
10410 int ret;
10411
10412 ret = print_trace_line(&iter);
10413 if (ret != TRACE_TYPE_NO_CONSUME)
10414 trace_consume(&iter);
10415 }
10416 touch_nmi_watchdog();
10417
10418 trace_printk_seq(&iter.seq);
10419 }
10420
10421 if (!cnt)
10422 printk(KERN_TRACE " (ftrace buffer empty)\n");
10423 else
10424 printk(KERN_TRACE "---------------------------------\n");
10425
10426 out_enable:
10427 tr->trace_flags |= old_userobj;
10428
10429 for_each_tracing_cpu(cpu) {
10430 atomic_dec(&per_cpu_ptr(iter.array_buffer->data, cpu)->disabled);
10431 }
10432 atomic_dec(&dump_running);
10433 local_irq_restore(flags);
10434 }
10435 EXPORT_SYMBOL_GPL(ftrace_dump);
10436
10437 #define WRITE_BUFSIZE 4096
10438
trace_parse_run_command(struct file * file,const char __user * buffer,size_t count,loff_t * ppos,int (* createfn)(const char *))10439 ssize_t trace_parse_run_command(struct file *file, const char __user *buffer,
10440 size_t count, loff_t *ppos,
10441 int (*createfn)(const char *))
10442 {
10443 char *kbuf, *buf, *tmp;
10444 int ret = 0;
10445 size_t done = 0;
10446 size_t size;
10447
10448 kbuf = kmalloc(WRITE_BUFSIZE, GFP_KERNEL);
10449 if (!kbuf)
10450 return -ENOMEM;
10451
10452 while (done < count) {
10453 size = count - done;
10454
10455 if (size >= WRITE_BUFSIZE)
10456 size = WRITE_BUFSIZE - 1;
10457
10458 if (copy_from_user(kbuf, buffer + done, size)) {
10459 ret = -EFAULT;
10460 goto out;
10461 }
10462 kbuf[size] = '\0';
10463 buf = kbuf;
10464 do {
10465 tmp = strchr(buf, '\n');
10466 if (tmp) {
10467 *tmp = '\0';
10468 size = tmp - buf + 1;
10469 } else {
10470 size = strlen(buf);
10471 if (done + size < count) {
10472 if (buf != kbuf)
10473 break;
10474 /* This can accept WRITE_BUFSIZE - 2 ('\n' + '\0') */
10475 pr_warn("Line length is too long: Should be less than %d\n",
10476 WRITE_BUFSIZE - 2);
10477 ret = -EINVAL;
10478 goto out;
10479 }
10480 }
10481 done += size;
10482
10483 /* Remove comments */
10484 tmp = strchr(buf, '#');
10485
10486 if (tmp)
10487 *tmp = '\0';
10488
10489 ret = createfn(buf);
10490 if (ret)
10491 goto out;
10492 buf += size;
10493
10494 } while (done < count);
10495 }
10496 ret = done;
10497
10498 out:
10499 kfree(kbuf);
10500
10501 return ret;
10502 }
10503
10504 #ifdef CONFIG_TRACER_MAX_TRACE
tr_needs_alloc_snapshot(const char * name)10505 __init static bool tr_needs_alloc_snapshot(const char *name)
10506 {
10507 char *test;
10508 int len = strlen(name);
10509 bool ret;
10510
10511 if (!boot_snapshot_index)
10512 return false;
10513
10514 if (strncmp(name, boot_snapshot_info, len) == 0 &&
10515 boot_snapshot_info[len] == '\t')
10516 return true;
10517
10518 test = kmalloc(strlen(name) + 3, GFP_KERNEL);
10519 if (!test)
10520 return false;
10521
10522 sprintf(test, "\t%s\t", name);
10523 ret = strstr(boot_snapshot_info, test) == NULL;
10524 kfree(test);
10525 return ret;
10526 }
10527
do_allocate_snapshot(const char * name)10528 __init static void do_allocate_snapshot(const char *name)
10529 {
10530 if (!tr_needs_alloc_snapshot(name))
10531 return;
10532
10533 /*
10534 * When allocate_snapshot is set, the next call to
10535 * allocate_trace_buffers() (called by trace_array_get_by_name())
10536 * will allocate the snapshot buffer. That will alse clear
10537 * this flag.
10538 */
10539 allocate_snapshot = true;
10540 }
10541 #else
do_allocate_snapshot(const char * name)10542 static inline void do_allocate_snapshot(const char *name) { }
10543 #endif
10544
enable_instances(void)10545 __init static void enable_instances(void)
10546 {
10547 struct trace_array *tr;
10548 char *curr_str;
10549 char *str;
10550 char *tok;
10551
10552 /* A tab is always appended */
10553 boot_instance_info[boot_instance_index - 1] = '\0';
10554 str = boot_instance_info;
10555
10556 while ((curr_str = strsep(&str, "\t"))) {
10557
10558 tok = strsep(&curr_str, ",");
10559
10560 if (IS_ENABLED(CONFIG_TRACER_MAX_TRACE))
10561 do_allocate_snapshot(tok);
10562
10563 tr = trace_array_get_by_name(tok, NULL);
10564 if (!tr) {
10565 pr_warn("Failed to create instance buffer %s\n", curr_str);
10566 continue;
10567 }
10568 /* Allow user space to delete it */
10569 trace_array_put(tr);
10570
10571 while ((tok = strsep(&curr_str, ","))) {
10572 early_enable_events(tr, tok, true);
10573 }
10574 }
10575 }
10576
tracer_alloc_buffers(void)10577 __init static int tracer_alloc_buffers(void)
10578 {
10579 int ring_buf_size;
10580 int ret = -ENOMEM;
10581
10582
10583 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10584 pr_warn("Tracing disabled due to lockdown\n");
10585 return -EPERM;
10586 }
10587
10588 /*
10589 * Make sure we don't accidentally add more trace options
10590 * than we have bits for.
10591 */
10592 BUILD_BUG_ON(TRACE_ITER_LAST_BIT > TRACE_FLAGS_MAX_SIZE);
10593
10594 if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
10595 goto out;
10596
10597 if (!alloc_cpumask_var(&global_trace.tracing_cpumask, GFP_KERNEL))
10598 goto out_free_buffer_mask;
10599
10600 /* Only allocate trace_printk buffers if a trace_printk exists */
10601 if (&__stop___trace_bprintk_fmt != &__start___trace_bprintk_fmt)
10602 /* Must be called before global_trace.buffer is allocated */
10603 trace_printk_init_buffers();
10604
10605 /* To save memory, keep the ring buffer size to its minimum */
10606 if (global_trace.ring_buffer_expanded)
10607 ring_buf_size = trace_buf_size;
10608 else
10609 ring_buf_size = 1;
10610
10611 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
10612 cpumask_copy(global_trace.tracing_cpumask, cpu_all_mask);
10613
10614 raw_spin_lock_init(&global_trace.start_lock);
10615
10616 /*
10617 * The prepare callbacks allocates some memory for the ring buffer. We
10618 * don't free the buffer if the CPU goes down. If we were to free
10619 * the buffer, then the user would lose any trace that was in the
10620 * buffer. The memory will be removed once the "instance" is removed.
10621 */
10622 ret = cpuhp_setup_state_multi(CPUHP_TRACE_RB_PREPARE,
10623 "trace/RB:prepare", trace_rb_cpu_prepare,
10624 NULL);
10625 if (ret < 0)
10626 goto out_free_cpumask;
10627 /* Used for event triggers */
10628 ret = -ENOMEM;
10629 temp_buffer = ring_buffer_alloc(PAGE_SIZE, RB_FL_OVERWRITE);
10630 if (!temp_buffer)
10631 goto out_rm_hp_state;
10632
10633 if (trace_create_savedcmd() < 0)
10634 goto out_free_temp_buffer;
10635
10636 if (!zalloc_cpumask_var(&global_trace.pipe_cpumask, GFP_KERNEL))
10637 goto out_free_savedcmd;
10638
10639 /* TODO: make the number of buffers hot pluggable with CPUS */
10640 if (allocate_trace_buffers(&global_trace, ring_buf_size) < 0) {
10641 MEM_FAIL(1, "tracer: failed to allocate ring buffer!\n");
10642 goto out_free_pipe_cpumask;
10643 }
10644 if (global_trace.buffer_disabled)
10645 tracing_off();
10646
10647 if (trace_boot_clock) {
10648 ret = tracing_set_clock(&global_trace, trace_boot_clock);
10649 if (ret < 0)
10650 pr_warn("Trace clock %s not defined, going back to default\n",
10651 trace_boot_clock);
10652 }
10653
10654 /*
10655 * register_tracer() might reference current_trace, so it
10656 * needs to be set before we register anything. This is
10657 * just a bootstrap of current_trace anyway.
10658 */
10659 global_trace.current_trace = &nop_trace;
10660
10661 global_trace.max_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
10662
10663 ftrace_init_global_array_ops(&global_trace);
10664
10665 init_trace_flags_index(&global_trace);
10666
10667 register_tracer(&nop_trace);
10668
10669 /* Function tracing may start here (via kernel command line) */
10670 init_function_trace();
10671
10672 /* All seems OK, enable tracing */
10673 tracing_disabled = 0;
10674
10675 atomic_notifier_chain_register(&panic_notifier_list,
10676 &trace_panic_notifier);
10677
10678 register_die_notifier(&trace_die_notifier);
10679
10680 global_trace.flags = TRACE_ARRAY_FL_GLOBAL;
10681
10682 INIT_LIST_HEAD(&global_trace.systems);
10683 INIT_LIST_HEAD(&global_trace.events);
10684 INIT_LIST_HEAD(&global_trace.hist_vars);
10685 INIT_LIST_HEAD(&global_trace.err_log);
10686 list_add(&global_trace.list, &ftrace_trace_arrays);
10687
10688 apply_trace_boot_options();
10689
10690 register_snapshot_cmd();
10691
10692 test_can_verify();
10693
10694 return 0;
10695
10696 out_free_pipe_cpumask:
10697 free_cpumask_var(global_trace.pipe_cpumask);
10698 out_free_savedcmd:
10699 free_saved_cmdlines_buffer(savedcmd);
10700 out_free_temp_buffer:
10701 ring_buffer_free(temp_buffer);
10702 out_rm_hp_state:
10703 cpuhp_remove_multi_state(CPUHP_TRACE_RB_PREPARE);
10704 out_free_cpumask:
10705 free_cpumask_var(global_trace.tracing_cpumask);
10706 out_free_buffer_mask:
10707 free_cpumask_var(tracing_buffer_mask);
10708 out:
10709 return ret;
10710 }
10711
ftrace_boot_snapshot(void)10712 void __init ftrace_boot_snapshot(void)
10713 {
10714 #ifdef CONFIG_TRACER_MAX_TRACE
10715 struct trace_array *tr;
10716
10717 if (!snapshot_at_boot)
10718 return;
10719
10720 list_for_each_entry(tr, &ftrace_trace_arrays, list) {
10721 if (!tr->allocated_snapshot)
10722 continue;
10723
10724 tracing_snapshot_instance(tr);
10725 trace_array_puts(tr, "** Boot snapshot taken **\n");
10726 }
10727 #endif
10728 }
10729
early_trace_init(void)10730 void __init early_trace_init(void)
10731 {
10732 if (tracepoint_printk) {
10733 tracepoint_print_iter =
10734 kzalloc(sizeof(*tracepoint_print_iter), GFP_KERNEL);
10735 if (MEM_FAIL(!tracepoint_print_iter,
10736 "Failed to allocate trace iterator\n"))
10737 tracepoint_printk = 0;
10738 else
10739 static_key_enable(&tracepoint_printk_key.key);
10740 }
10741 tracer_alloc_buffers();
10742
10743 init_events();
10744 }
10745
trace_init(void)10746 void __init trace_init(void)
10747 {
10748 trace_event_init();
10749
10750 if (boot_instance_index)
10751 enable_instances();
10752 }
10753
clear_boot_tracer(void)10754 __init static void clear_boot_tracer(void)
10755 {
10756 /*
10757 * The default tracer at boot buffer is an init section.
10758 * This function is called in lateinit. If we did not
10759 * find the boot tracer, then clear it out, to prevent
10760 * later registration from accessing the buffer that is
10761 * about to be freed.
10762 */
10763 if (!default_bootup_tracer)
10764 return;
10765
10766 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
10767 default_bootup_tracer);
10768 default_bootup_tracer = NULL;
10769 }
10770
10771 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
tracing_set_default_clock(void)10772 __init static void tracing_set_default_clock(void)
10773 {
10774 /* sched_clock_stable() is determined in late_initcall */
10775 if (!trace_boot_clock && !sched_clock_stable()) {
10776 if (security_locked_down(LOCKDOWN_TRACEFS)) {
10777 pr_warn("Can not set tracing clock due to lockdown\n");
10778 return;
10779 }
10780
10781 printk(KERN_WARNING
10782 "Unstable clock detected, switching default tracing clock to \"global\"\n"
10783 "If you want to keep using the local clock, then add:\n"
10784 " \"trace_clock=local\"\n"
10785 "on the kernel command line\n");
10786 tracing_set_clock(&global_trace, "global");
10787 }
10788 }
10789 #else
tracing_set_default_clock(void)10790 static inline void tracing_set_default_clock(void) { }
10791 #endif
10792
late_trace_init(void)10793 __init static int late_trace_init(void)
10794 {
10795 if (tracepoint_printk && tracepoint_printk_stop_on_boot) {
10796 static_key_disable(&tracepoint_printk_key.key);
10797 tracepoint_printk = 0;
10798 }
10799
10800 tracing_set_default_clock();
10801 clear_boot_tracer();
10802 return 0;
10803 }
10804
10805 late_initcall_sync(late_trace_init);
10806